id
int64 1
24.5M
| paragraph_id
int64 0
159k
| content
stringlengths 1
571k
⌀ | paper_arxiv_id
stringlengths 10
13
| paper_section
stringlengths 1
262
| section_id
float64 4
360k
⌀ | paragraph_in_paper_id
int64 1
29.8k
|
|---|---|---|---|---|---|---|
22,511,843
| 39
|
\begin{property}[Easy Gas Estimation]\label{prop:Composability} Given two sets of transactions $\txset_1$ and $\txset_2$ and a transaction $\tx$ belonging to neither $T_1$ nor $T_2$:
\[
\gas_{\txset_1 \cup \{\tx\}}(\tx) =
\gas_{\txset_2 \cup \{\tx\}}(\tx).
\]
\end{property}\begin{property}[Easy Gas Estimation]\label{prop:Composability} Given two sets of transactions $\txset_1$ and $\txset_2$ and a transaction $\tx$ belonging to neither $T_1$ nor $T_2$:
\[
\gas_{\txset_1 \cup \{\tx\}}(\tx) =
\gas_{\txset_2 \cup \{\tx\}}(\tx).
\]
\end{property}[Easy Gas Estimation]\label{prop:Composability} Given two sets of transactions $\txset_1$\txset_1 and $\txset_2$\txset_2 and a transaction $\tx$\tx belonging to neither $T_1$T_1 nor $T_2$T_2:
\[
\gas_{\txset_1 \cup \{\tx\}}(\tx) =
\gas_{\txset_2 \cup \{\tx\}}(\tx).
\]
\gas_{\txset_1 \cup \{\tx\}}\txset_1 \cup \{\tx\}(\tx) =
\gas_{\txset_2 \cup \{\tx\}}\txset_2 \cup \{\tx\}(\tx).
|
2502.11964
|
Properties
| 228,271
| 60
|
22,511,844
| 40
|
This property ensures a good user experience by making it easy for users to estimate gas usage. If the gas consumption of a transaction depended on the remaining transactions in the block, this estimation would become significantly more complex. In general, we aim to keep gas estimation straightforward to avoid giving an advantage to more sophisticated users. Additionally, as we will see later, a \GCM satisfying this property can be seamlessly composed with existing \TFMs, retaining their desirable properties (see \cref{sec:outlook}). Sadly, as one might have already guessed, Easy Gas Estimation is incompatible with Scheduling Monotonicity (except for \emph{constant} mechanisms, i.e., \GCMs that return a constant independent of $T$T and $\tx$\tx), and also incompatible with Efficiency:
|
2502.11964
|
Properties
| 228,271
| 61
|
22,511,845
| 41
|
\begin{theorem}\label{thm:impossibility} Easy Gas Estimation (\cref{prop:Composability}) is:
%
\begin{enumerate}
\item Incompatible with Scheduling Monotonicity (\cref{prop:schedule_monotonicity}) unless using a constant \GCM.\footnote{Constant \GCMs satisfy Scheduling Monotonicity, but not Strict Scheduling Monotonicity.}
\item Incompatible with Efficiency (\cref{prop:efficiency}).
\end{enumerate}
\end{theorem}\begin{theorem}\label{thm:impossibility} Easy Gas Estimation (\cref{prop:Composability}) is:
%
\begin{enumerate}
\item Incompatible with Scheduling Monotonicity (\cref{prop:schedule_monotonicity}) unless using a constant \GCM.\footnote{Constant \GCMs satisfy Scheduling Monotonicity, but not Strict Scheduling Monotonicity.}
\item Incompatible with Efficiency (\cref{prop:efficiency}).
\end{enumerate}
\end{theorem}\label{thm:impossibility} Easy Gas Estimation (\cref{prop:Composability}) is:
|
2502.11964
|
Properties
| 228,271
| 62
|
22,511,846
| 42
|
\item Incompatible with Scheduling Monotonicity (\cref{prop:schedule_monotonicity}) unless using a constant \GCM.\footnote{Constant \GCMs satisfy Scheduling Monotonicity, but not Strict Scheduling Monotonicity.}
\item Incompatible with Efficiency (\cref{prop:efficiency}).
|
2502.11964
|
Properties
| 228,271
| 63
|
22,511,848
| 44
|
Consider a mechanism $M$ with Easy Gas Estimation; i.e., $\gas^M(\tx)$ is meaningful without a subscript for the set of transactions $T$ in the block. Then:
|
2502.11964
|
Properties
| 228,271
| 65
|
22,511,849
| 45
|
\begin{enumerate}[topsep=0pt,itemsep=-1ex,partopsep=1ex,parsep=1ex]
\item Assume that $M$ has Scheduling Monotonicity; we will show that $M$ is a constant mechanism.
|
2502.11964
|
Properties
| 228,271
| 66
|
22,511,850
| 46
|
Call two transactions $\tx_1 \simeq (t_1, K_1)$ and $\tx_2 \simeq (t_2, K_2)$ \emph{incomparable} if neither $K_1 \subseteq K_2$ nor $K_2 \subseteq K_1$. As a first step, we will show that if $\tx_1$ and $\tx_2$ are incomparable, then $\gas^M(\tx_1) = \gas^M(\tx_2)$. To show this, assume that $K_1 \not\subseteq K_2$. We will show that then $\gas^M(\tx_1) \geq \gas^M(\tx_2)$. Exchanging the roles of $\tx_1$ and $\tx_2$ will then give the conclusion. Let $k \in K_1 \setminus K_2$ and $\tx_3$ be another transaction such that $\tx_3 \simeq (t_3, \{k\})$ for some $t_3 > t_2$, and define $T = \{\tx_3\}$. Then, for any number of threads $n \geq 2$ and any scheduler that is optimal for two transactions, we have $v(T \cup \{\tx_1\}) = t_1 + t_3 > t_3 = v(T \cup \{\tx_2\})$. By Scheduling Monotonicity, this implies that $\gas^M(\tx_1) \geq \gas^M(\tx_2)$.
We now know that $M$ associates the same gas consumption to any pair of incomparable transactions. Armed as such, consider two arbitrary transactions $\tx_1 \simeq (t_1, K_1)$ and $\tx_2 \simeq (t_2, K_2)$. Let $k$ be an arbitrary storage key \emph{not} in $K_1 \cup K_2$ and $\tx_3$ be a transaction such that $\tx_3 \simeq (1, \{r\})$. Since $K_1$ and $K_2$ are non-empty, one can easily see that $\tx_3$ is incomparable with both $\tx_1$ and $\tx_2$, from which $\gas^M(\tx_1) = \gas^M(\tx_3) = \gas^M(\tx_2)$.
|
2502.11964
|
Properties
| 228,271
| 67
|
22,511,851
| 47
|
\item Assume for a contradiction that $M$ has Efficiency, then for any transaction $\tx \simeq (\txtime, \txkeys)$ we have $\gas^M(\tx) = \gas_\varnothing^M(\{\tx\}) = v(\{\tx\}) = \txtime$. Hence, by definition, $M$ is the current mechanism, which can be easily seen to not satisfy Efficiency: assume $n \geq 2$ threads and consider the set of transactions $T = \{\tx_1, \tx_2\}$ where $\tx_1 \simeq (1, \{k_1\})$ and $\tx_2 \simeq (1, \{k_2\})$. In this case, $\gas^M(T) = \gas^M(\tx_1) + \gas^M(\tx_1) = 1 + 1 = 2 \neq 1 = v(T)$. \qedhere
\end{enumerate}
\end{proof}\begin{proof}
|
2502.11964
|
Properties
| 228,271
| 68
|
22,511,852
| 48
|
Consider a mechanism $M$ with Easy Gas Estimation; i.e., $\gas^M(\tx)$ is meaningful without a subscript for the set of transactions $T$ in the block. Then:
|
2502.11964
|
Properties
| 228,271
| 69
|
22,511,853
| 49
|
\begin{enumerate}[topsep=0pt,itemsep=-1ex,partopsep=1ex,parsep=1ex]
\item Assume that $M$ has Scheduling Monotonicity; we will show that $M$ is a constant mechanism.
|
2502.11964
|
Properties
| 228,271
| 70
|
22,511,854
| 50
|
Call two transactions $\tx_1 \simeq (t_1, K_1)$ and $\tx_2 \simeq (t_2, K_2)$ \emph{incomparable} if neither $K_1 \subseteq K_2$ nor $K_2 \subseteq K_1$. As a first step, we will show that if $\tx_1$ and $\tx_2$ are incomparable, then $\gas^M(\tx_1) = \gas^M(\tx_2)$. To show this, assume that $K_1 \not\subseteq K_2$. We will show that then $\gas^M(\tx_1) \geq \gas^M(\tx_2)$. Exchanging the roles of $\tx_1$ and $\tx_2$ will then give the conclusion. Let $k \in K_1 \setminus K_2$ and $\tx_3$ be another transaction such that $\tx_3 \simeq (t_3, \{k\})$ for some $t_3 > t_2$, and define $T = \{\tx_3\}$. Then, for any number of threads $n \geq 2$ and any scheduler that is optimal for two transactions, we have $v(T \cup \{\tx_1\}) = t_1 + t_3 > t_3 = v(T \cup \{\tx_2\})$. By Scheduling Monotonicity, this implies that $\gas^M(\tx_1) \geq \gas^M(\tx_2)$.
We now know that $M$ associates the same gas consumption to any pair of incomparable transactions. Armed as such, consider two arbitrary transactions $\tx_1 \simeq (t_1, K_1)$ and $\tx_2 \simeq (t_2, K_2)$. Let $k$ be an arbitrary storage key \emph{not} in $K_1 \cup K_2$ and $\tx_3$ be a transaction such that $\tx_3 \simeq (1, \{r\})$. Since $K_1$ and $K_2$ are non-empty, one can easily see that $\tx_3$ is incomparable with both $\tx_1$ and $\tx_2$, from which $\gas^M(\tx_1) = \gas^M(\tx_3) = \gas^M(\tx_2)$.
|
2502.11964
|
Properties
| 228,271
| 71
|
22,511,858
| 54
|
Call two transactions $\tx_1 \simeq (t_1, K_1)$\tx_1 \simeq (t_1, K_1) and $\tx_2 \simeq (t_2, K_2)$\tx_2 \simeq (t_2, K_2) \emph{incomparable} if neither $K_1 \subseteq K_2$K_1 \subseteq K_2 nor $K_2 \subseteq K_1$K_2 \subseteq K_1. As a first step, we will show that if $\tx_1$\tx_1 and $\tx_2$\tx_2 are incomparable, then $\gas^M(\tx_1) = \gas^M(\tx_2)$\gas^M(\tx_1) = \gas^M(\tx_2). To show this, assume that $K_1 \not\subseteq K_2$K_1 \not\subseteq K_2. We will show that then $\gas^M(\tx_1) \geq \gas^M(\tx_2)$\gas^M(\tx_1) \geq \gas^M(\tx_2). Exchanging the roles of $\tx_1$\tx_1 and $\tx_2$\tx_2 will then give the conclusion. Let $k \in K_1 \setminus K_2$k \in K_1 \setminus K_2 and $\tx_3$\tx_3 be another transaction such that $\tx_3 \simeq (t_3, \{k\})$\tx_3 \simeq (t_3, \{k\}) for some $t_3 > t_2$t_3 > t_2, and define $T = \{\tx_3\}$T = \{\tx_3\}. Then, for any number of threads $n \geq 2$n \geq 2 and any scheduler that is optimal for two transactions, we have $v(T \cup \{\tx_1\}) = t_1 + t_3 > t_3 = v(T \cup \{\tx_2\})$v(T \cup \{\tx_1\}) = t_1 + t_3 > t_3 = v(T \cup \{\tx_2\}). By Scheduling Monotonicity, this implies that $\gas^M(\tx_1) \geq \gas^M(\tx_2)$\gas^M(\tx_1) \geq \gas^M(\tx_2).
We now know that $M$M associates the same gas consumption to any pair of incomparable transactions. Armed as such, consider two arbitrary transactions $\tx_1 \simeq (t_1, K_1)$\tx_1 \simeq (t_1, K_1) and $\tx_2 \simeq (t_2, K_2)$\tx_2 \simeq (t_2, K_2). Let $k$k be an arbitrary storage key \emph{not} in $K_1 \cup K_2$K_1 \cup K_2 and $\tx_3$\tx_3 be a transaction such that $\tx_3 \simeq (1, \{r\})$\tx_3 \simeq (1, \{r\}). Since $K_1$K_1 and $K_2$K_2 are non-empty, one can easily see that $\tx_3$\tx_3 is incomparable with both $\tx_1$\tx_1 and $\tx_2$\tx_2, from which $\gas^M(\tx_1) = \gas^M(\tx_3) = \gas^M(\tx_2)$\gas^M(\tx_1) = \gas^M(\tx_3) = \gas^M(\tx_2).
|
2502.11964
|
Properties
| 228,271
| 75
|
22,511,859
| 55
|
\item Assume for a contradiction that $M$M has Efficiency, then for any transaction $\tx \simeq (\txtime, \txkeys)$\tx \simeq (\txtime, \txkeys) we have $\gas^M(\tx) = \gas_\varnothing^M(\{\tx\}) = v(\{\tx\}) = \txtime$\gas^M(\tx) = \gas_\varnothing^M(\{\tx\}) = v(\{\tx\}) = \txtime. Hence, by definition, $M$M is the current mechanism, which can be easily seen to not satisfy Efficiency: assume $n \geq 2$n \geq 2 threads and consider the set of transactions $T = \{\tx_1, \tx_2\}$T = \{\tx_1, \tx_2\} where $\tx_1 \simeq (1, \{k_1\})$\tx_1 \simeq (1, \{k_1\}) and $\tx_2 \simeq (1, \{k_2\})$\tx_2 \simeq (1, \{k_2\}). In this case, $\gas^M(T) = \gas^M(\tx_1) + \gas^M(\tx_1) = 1 + 1 = 2 \neq 1 = v(T)$\gas^M(T) = \gas^M(\tx_1) + \gas^M(\tx_1) = 1 + 1 = 2 \neq 1 = v(T). \qedhere
|
2502.11964
|
Properties
| 228,271
| 76
|
22,511,860
| 56
|
Our second practical property requires that gas consumption be efficiently computable, i.e., in polynomial time (\cref{prop:computational_complexity}). A \GCM that does not satisfy this property would be unsuitable for the blockchain context, where gas computation is intended to be a straightforward component.
|
2502.11964
|
Properties
| 228,271
| 77
|
22,511,861
| 57
|
\begin{property}[Poly-time Computable]\label{prop:computational_complexity}
There exists a polynomial-time algorithm that takes as input a transaction set $\txset$ and a transaction $\tx$ and outputs
$\gas_{\txset \cup \{\tx\}}(\tx)$.
\end{property}\begin{property}[Poly-time Computable]\label{prop:computational_complexity}
There exists a polynomial-time algorithm that takes as input a transaction set $\txset$ and a transaction $\tx$ and outputs
$\gas_{\txset \cup \{\tx\}}(\tx)$.
\end{property}[Poly-time Computable]\label{prop:computational_complexity}
There exists a polynomial-time algorithm that takes as input a transaction set $\txset$\txset and a transaction $\tx$\tx and outputs
$\gas_{\txset \cup \{\tx\}}(\tx)$\gas_{\txset \cup \{\tx\}}\txset \cup \{\tx\}(\tx).
|
2502.11964
|
Properties
| 228,271
| 78
|
22,511,863
| 1
|
We now propose multiple \GCM designs. Motivated by the incompatibilities in \cref{thm:impossibility}, these fall into two categories:
|
2502.11964
|
Proposals
| 228,272
| 80
|
22,511,864
| 2
|
\begin{enumerate}[label=(C\arabic*),topsep=0pt,itemsep=-1ex,partopsep=1ex,parsep=1ex,leftmargin=2.45em]
\item Mechanisms with Easy Gas Estimation, in which each transaction's gas consumption is computed in isolation. Such mechanisms tend to be both straightforward and attractive but can achieve neither Scheduling Monotonicity\footnote{Except for constant \GCMs.} nor Efficiency.
\item Mechanisms \emph{without} Easy Gas Estimation, which, given a set of transactions $\txset$, rely on $v(T)$ or, more generally, $\left(v(\txset')\right)_{\txset' \subseteq \txset'}$ to holistically calculate gas consumptions for the block $\txset$, requiring knowledge of the entire block. Instead, these mechanisms aim to achieve Scheduling Monotonicity and/or Efficiency.
\end{enumerate}\begin{enumerate}[label=(C\arabic*),topsep=0pt,itemsep=-1ex,partopsep=1ex,parsep=1ex,leftmargin=2.45em]
\item Mechanisms with Easy Gas Estimation, in which each transaction's gas consumption is computed in isolation. Such mechanisms tend to be both straightforward and attractive but can achieve neither Scheduling Monotonicity\footnote{Except for constant \GCMs.} nor Efficiency.
\item Mechanisms \emph{without} Easy Gas Estimation, which, given a set of transactions $\txset$, rely on $v(T)$ or, more generally, $\left(v(\txset')\right)_{\txset' \subseteq \txset'}$ to holistically calculate gas consumptions for the block $\txset$, requiring knowledge of the entire block. Instead, these mechanisms aim to achieve Scheduling Monotonicity and/or Efficiency.
\end{enumerate}
\item Mechanisms with Easy Gas Estimation, in which each transaction's gas consumption is computed in isolation. Such mechanisms tend to be both straightforward and attractive but can achieve neither Scheduling Monotonicity\footnote{Except for constant \GCMs.} nor Efficiency.
\item Mechanisms \emph{without} Easy Gas Estimation, which, given a set of transactions $\txset$\txset, rely on $v(T)$v(T) or, more generally, $\left(v(\txset')\right)_{\txset' \subseteq \txset'}$\left(v(\txset')\right)_{\txset' \subseteq \txset'}\txset' \subseteq \txset' to holistically calculate gas consumptions for the block $\txset$\txset, requiring knowledge of the entire block. Instead, these mechanisms aim to achieve Scheduling Monotonicity and/or Efficiency.
|
2502.11964
|
Proposals
| 228,272
| 81
|
22,511,866
| 4
|
We already introduced the \emph{current} \GCM (\cref{def:current-gas-comp-mech}). For the next mechanism, we assume a globally available vector of \emph{positive} weights for the storage keys $(w_k)_{k \in \keyset}$(w_k)_{k \in \keyset}k \in \keyset.
For instance, these weights could all be 1. Alternatively, higher-weight storage keys could correspond to storage keys under higher (historical) demand. For our purposes, we only need to assume that the weights are known and do not depend on the block being built. In practice, one could update the weights between blocks to accurately reflect storage key demand (the exact details are not relevant here; see \cref{sec:outlook} for further discussion). Given the weights, we define:
|
2502.11964
|
Proposals
| 228,272
| 83
|
22,511,867
| 5
|
\begin{definition}[Weighted Area \GCM]\label{def:weighted-area-gas-comp-mech}
Given a set of transactions $\txset$ and a transaction $\tx \in T$ with $\tx \simeq (t, K)$, the \emph{Weighted Area} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation}
\gas_{\txset}^\textit{WA}(\tx) := t \cdot \left(1 + \sum_{k \in K} w_k\right)
\label{def:weighted-area-eq}
\end{equation}
Since this does not depend on $T$, we often drop the subscript.
\end{definition}\begin{definition}[Weighted Area \GCM]\label{def:weighted-area-gas-comp-mech}
Given a set of transactions $\txset$ and a transaction $\tx \in T$ with $\tx \simeq (t, K)$, the \emph{Weighted Area} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation}
\gas_{\txset}^\textit{WA}(\tx) := t \cdot \left(1 + \sum_{k \in K} w_k\right)
\label{def:weighted-area-eq}
\end{equation}
Since this does not depend on $T$, we often drop the subscript.
\end{definition}\label{def:weighted-area-gas-comp-mech}
Given a set of transactions $\txset$\txset and a transaction $\tx \in T$\tx \in T with $\tx \simeq (t, K)$\tx \simeq (t, K), the \emph{Weighted Area} \GCM computes the amount of gas used by $\tx$\tx as follows:
\gas_{\txset}\txset^\textit{WA}(\tx) := t \cdot \left(1 + \sum_{k \in K}k \in K w_k\right)
\label{def:weighted-area-eq}
Since this does not depend on $T$T, we often drop the subscript.
\vspace{2pt}
|
2502.11964
|
Proposals
| 228,272
| 84
|
22,511,876
| 14
|
\begin{lemma}\label{lemma:shapley-efficient} The Shapley \GCM satisfies Efficiency (\cref{prop:efficiency}).
\end{lemma}\begin{lemma}\label{lemma:shapley-efficient} The Shapley \GCM satisfies Efficiency (\cref{prop:efficiency}).
\end{lemma}\label{lemma:shapley-efficient} The Shapley \GCM satisfies Efficiency (\cref{prop:efficiency}).
|
2502.11964
|
Proposals
| 228,272
| 93
|
22,511,877
| 15
|
We postpone studying the Shapley \GCM further until introducing our other mechanisms in the section.
|
2502.11964
|
Proposals
| 228,272
| 94
|
22,511,868
| 6
|
To gain intuition, it is instructive to consider the former case, where all weights are 1, i.e., the \emph{unweighted area} mechanism, in which case \cref{def:weighted-area-eq} becomes $\gas_{\txset}^\textit{WA}(\tx) = \txtime \cdot \left(1 + |\txkeys|\right)$\gas_{\txset}\txset^\textit{WA}(\tx) = \txtime \cdot \left(1 + |\txkeys|\right).
This mechanism can also be viewed as the addition of two terms: the \emph{current term}, i.e., the gas consumption of $\tx$\tx under the \emph{current} \GCM, namely $t$t, and the \emph{area term}, i.e., the area of a $t \times |K|$t \times |K| rectangle (which is also how we draw transactions in our diagrams). The \emph{current term} is helpful to ensure no transaction ever costs too little when the weights of all storage keys accessed by a transaction are too close to zero.\footnote{This term has a practical motivation. In reality, transactions will be scheduled on a small finite number of threads. Thus, even if they do not access any high-demand storage key, they occupy space in the schedule proportional to their execution time. Note, moreover, that this term could technically be omitted, and our results would still hold.}
|
2502.11964
|
Proposals
| 228,272
| 85
|
22,511,869
| 7
|
The \emph{area term}, on the other hand, which is the main component, intuitively corresponds to ``negated throughput.'' That is, executing the transaction requires holding $|K|$|K| locks for $t$t time units, during which no other transactions requiring any of those storage keys can execute. In \cref{fig:area}, we illustrate this idea. Transaction $\tx$\tx prevents the execution of other transactions across its entire storage key set but only utilizes each storage key for a short period. While the current \GCM charges a transaction based solely on its total computation time (i.e., the height of the rectangle), the weighted area \GCM also accounts for the storage keys for which it ``negated throughput'', i.e., the area occupied by the transaction. This occupied area prevents other transactions using the same storage keys from being scheduled in parallel.
\noindent
\begin{minipage}[b]{0.42\textwidth}
\quad Along the same line of reasoning, to further reinforce why this mechanism is a reasonable choice, consider a transaction set $T$. The sum $\sum_{\tx \in T} t(\tx) \cdot |\txkeys(\tx)|$ represents the ``total area'' of transactions in $\txset$. Dividing this term by the number of storage keys used in $\txset$ provides a lower bound on $v(\txset)$, serving as a rough proxy that does not require knowledge of $T$ when computing individual gas consumptions. Finally, we note that, even in the case of non-unit weights, the term \emph{weighted area} remains meaningful, as the area term still corresponds to the area of the respective rectangle in our diagrams if we give the column of each storage key $k \in \keyset$ a width of $w_k$.
\end{minipage}[b]{0.42\textwidth}0.42\textwidth
\quad Along the same line of reasoning, to further reinforce why this mechanism is a reasonable choice, consider a transaction set $T$T. The sum $\sum_{\tx \in T} t(\tx) \cdot |\txkeys(\tx)|$\sum_{\tx \in T}\tx \in T t(\tx) \cdot |\txkeys(\tx)| represents the ``total area'' of transactions in $\txset$\txset. Dividing this term by the number of storage keys used in $\txset$\txset provides a lower bound on $v(\txset)$v(\txset), serving as a rough proxy that does not require knowledge of $T$T when computing individual gas consumptions. Finally, we note that, even in the case of non-unit weights, the term \emph{weighted area} remains meaningful, as the area term still corresponds to the area of the respective rectangle in our diagrams if we give the column of each storage key $k \in \keyset$k \in \keyset a width of $w_k$w_k.
\hfill
\begin{minipage}[b]{0.5\textwidth}
\centering
\vspace{-8pt}
\input{tikz/area_a.tikz}\vspace{-10pt}
\captionof{figure}{Illustration of a worst-case transaction in terms of parallelizability. Transaction $\tx \simeq (6,\{k_2,\dots,k_7\})$ (shown in \fcolorbox{black}{color1!20}{\rule{0pt}{3pt}\rule{3pt}{0pt}}\hspace{1pt}) takes 6 units of time to execute and accesses 6 storage keys. However, the transaction only operates on each storage key for one unit of time each (shown in \fcolorbox{black}{color1!50}{\rule{0pt}{3pt}\rule{3pt}{0pt}}\hspace{1pt}).}
\label{fig:area}\vspace{0pt}
\end{minipage}[b]{0.5\textwidth}0.5\textwidth
\centering
\vspace{-8pt}
\input{tikz/area_a.tikz}\vspace{-10pt}
\captionof{figure}figure{Illustration of a worst-case transaction in terms of parallelizability. Transaction $\tx \simeq (6,\{k_2,\dots,k_7\})$ (shown in \fcolorbox{black}{color1!20}{\rule{0pt}{3pt}\rule{3pt}{0pt}}\hspace{1pt}) takes 6 units of time to execute and accesses 6 storage keys. However, the transaction only operates on each storage key for one unit of time each (shown in \fcolorbox{black}{color1!50}{\rule{0pt}{3pt}\rule{3pt}{0pt}}\hspace{1pt}).}Illustration of a worst-case transaction in terms of parallelizability. Transaction $\tx \simeq (6,\{k_2,\dots,k_7\})$\tx \simeq (6,\{k_2,\dots,k_7\}) (shown in \fcolorbox{black}{color1!20}{\rule{0pt}{3pt}\rule{3pt}{0pt}}\hspace{1pt}) takes 6 units of time to execute and accesses 6 storage keys. However, the transaction only operates on each storage key for one unit of time each (shown in \fcolorbox{black}{color1!50}{\rule{0pt}{3pt}\rule{3pt}{0pt}}\hspace{1pt}).
\label{fig:area}\vspace{0pt}
|
2502.11964
|
Proposals
| 228,272
| 86
|
22,511,871
| 9
|
We now switch gears towards mechanisms without Easy Gas Estimation that instead aim to achieve Scheduling Monotonicity and/or Efficiency. Intuitively, for a set of transactions $\txset$\txset, such mechanisms should start from $v(T)$v(T) and take each transaction's marginal contribution towards $v(T)$v(T) as its gas consumption. This, however, has to be done carefully, as, e.g., simply looking for each transaction $\tx \in \txset$\tx \in \txset at $v(\txset) - v(\txset \setminus \{\tx\})$v(\txset) - v(\txset \setminus \{\tx\}) does not help. Note that one can easily construct cases where removing any one transaction from $\txset$\txset does not change $v(\txset)$v(\txset), so all reported numbers will be 0. However, all these numbers being zero does not mean that no transaction contributes to $v(\txset)$v(\txset); it just means that we also have to consider removing multiple transactions to see the differences. This gives us the idea to look at the marginal contribution of each transaction $\tx \in \txset$\tx \in \txset when added to each possible subset $S\subseteq \txset\setminus\{\tx\}$S\subseteq \txset\setminus\{\tx\}, and not just to $S = \txset\setminus\{\tx\}$S = \txset\setminus\{\tx\} as before. Formally, we would like to compute $\tx$\tx's gas consumption as an aggregate of $\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}$\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}S\subseteq \txset\setminus\{\tx\}. We next present two mechanisms based on this idea: the \emph{Shapley} (\cref{def:shapley-gas-comp-mech}) and \emph{Banzhaf} (\cref{def:banzhaf-gas-comp-mech}) \GCMs, inspired by corresponding concepts in cooperative game theory.
|
2502.11964
|
Proposals
| 228,272
| 88
|
22,511,872
| 10
|
\begin{definition}[Shapley \GCM]\label{def:shapley-gas-comp-mech}
Given a set of transactions $\txset$
consisting of $|\txset| = n$ transactions
and a transaction $\tx \in \txset$, the \emph{Shapley} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{align}
\gas_{\txset}^S(\tx) := & \frac{1}{n!} \sum_\sigma \left[ v(P_{\tx}^\sigma \cup \{\tx\}) - v(P_{\tx}^\sigma) \right] \label{shapley:v1}\\
= & \sum_{S\subseteq \txset\setminus\{\tx\}} \frac { |S|!\cdot (n-|S|-1)!} { n! } \left[ v ( S\cup\{ \tx\} ) - v(S) \right] \label{shapley:v2}.
\end{align}
Here, $\sigma$ ranges over the $n!$ possible ways to order the transactions in $\txset$ and $P_{\tx}^\sigma$ denotes the set of transactions that precede $\tx$ in the order $\sigma$. The equality between \cref{shapley:v1,shapley:v2} follows by counting the number of orders $\sigma$ such that $P_{\tx}^\sigma = S$, which there are $|S|! \cdot (n-|S|-1)!$ of.
\end{definition}\begin{definition}[Shapley \GCM]\label{def:shapley-gas-comp-mech}
Given a set of transactions $\txset$
consisting of $|\txset| = n$ transactions
and a transaction $\tx \in \txset$, the \emph{Shapley} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{align}
\gas_{\txset}^S(\tx) := & \frac{1}{n!} \sum_\sigma \left[ v(P_{\tx}^\sigma \cup \{\tx\}) - v(P_{\tx}^\sigma) \right] \label{shapley:v1}\\
= & \sum_{S\subseteq \txset\setminus\{\tx\}} \frac { |S|!\cdot (n-|S|-1)!} { n! } \left[ v ( S\cup\{ \tx\} ) - v(S) \right] \label{shapley:v2}.
\end{align}
Here, $\sigma$ ranges over the $n!$ possible ways to order the transactions in $\txset$ and $P_{\tx}^\sigma$ denotes the set of transactions that precede $\tx$ in the order $\sigma$. The equality between \cref{shapley:v1,shapley:v2} follows by counting the number of orders $\sigma$ such that $P_{\tx}^\sigma = S$, which there are $|S|! \cdot (n-|S|-1)!$ of.
\end{definition}\label{def:shapley-gas-comp-mech}
Given a set of transactions $\txset$\txset
consisting of $|\txset| = n$|\txset| = n transactions
and a transaction $\tx \in \txset$\tx \in \txset, the \emph{Shapley} \GCM computes the amount of gas used by $\tx$\tx as follows:
\gas_{\txset}\txset^S(\tx) := & \frac{1}{n!} \sum_\sigma \left[ v(P_{\tx}\tx^\sigma \cup \{\tx\}) - v(P_{\tx}\tx^\sigma) \right] \label{shapley:v1}\\
= & \sum_{S\subseteq \txset\setminus\{\tx\}}S\subseteq \txset\setminus\{\tx\} \frac { |S|!\cdot (n-|S|-1)!} { n! } \left[ v ( S\cup\{ \tx\} ) - v(S) \right] \label{shapley:v2}.
Here, $\sigma$\sigma ranges over the $n!$n! possible ways to order the transactions in $\txset$\txset and $P_{\tx}^\sigma$P_{\tx}\tx^\sigma denotes the set of transactions that precede $\tx$\tx in the order $\sigma$\sigma. The equality between \cref{shapley:v1,shapley:v2} follows by counting the number of orders $\sigma$\sigma such that $P_{\tx}^\sigma = S$P_{\tx}\tx^\sigma = S, which there are $|S|! \cdot (n-|S|-1)!$|S|! \cdot (n-|S|-1)! of.
\vspace{2pt}
|
2502.11964
|
Proposals
| 228,272
| 89
|
22,511,873
| 11
|
Another way to understand the Shapley \GCM is through the following probabilistic experiment:
|
2502.11964
|
Proposals
| 228,272
| 90
|
22,511,874
| 12
|
\begin{enumerate}
\item Select an ordering $\sigma$ of the transactions in $\txset$ uniformly at random.
\item Start with an empty set of transactions and add transactions one by one in the order given by $\sigma$.
\item Whenever a transaction $\tx$ is added, let $S = P_{\tx}^\sigma$ be the set of transactions just before adding it and compute $\tx$'s \emph{marginal contribution} to the execution time as $v(S \cup \{tx\}) - v(S)$; i.e., by how much did the execution time increase by adding $\tx$ to the current set of transactions.
\item The gas consumption of $\tx \in \txset$ is the expectation of its marginal contribution across the experiment.
\end{enumerate}\begin{enumerate}
\item Select an ordering $\sigma$ of the transactions in $\txset$ uniformly at random.
\item Start with an empty set of transactions and add transactions one by one in the order given by $\sigma$.
\item Whenever a transaction $\tx$ is added, let $S = P_{\tx}^\sigma$ be the set of transactions just before adding it and compute $\tx$'s \emph{marginal contribution} to the execution time as $v(S \cup \{tx\}) - v(S)$; i.e., by how much did the execution time increase by adding $\tx$ to the current set of transactions.
\item The gas consumption of $\tx \in \txset$ is the expectation of its marginal contribution across the experiment.
\end{enumerate}
\item Select an ordering $\sigma$\sigma of the transactions in $\txset$\txset uniformly at random.
\item Start with an empty set of transactions and add transactions one by one in the order given by $\sigma$\sigma.
\item Whenever a transaction $\tx$\tx is added, let $S = P_{\tx}^\sigma$S = P_{\tx}\tx^\sigma be the set of transactions just before adding it and compute $\tx$\tx's \emph{marginal contribution} to the execution time as $v(S \cup \{tx\}) - v(S)$v(S \cup \{tx\}) - v(S); i.e., by how much did the execution time increase by adding $\tx$\tx to the current set of transactions.
\item The gas consumption of $\tx \in \txset$\tx \in \txset is the expectation of its marginal contribution across the experiment.
|
2502.11964
|
Proposals
| 228,272
| 91
|
22,511,875
| 13
|
The reader familiar with cooperative game theory will have already recognized the immediate connection with Shapley values: if we see each transaction $\tx \in \txset$\tx \in \txset as a player in a game with valuation function $v : 2^\txset \to \mathbb{R}$v : 2^\txset \to \mathbb{R}, then $\gas_{\txset}^S(\tx)$\gas_{\txset}\txset^S(\tx) is precisely the celebrated \emph{Shapley value} of player $\tx$\tx, more traditionally written $\phi_{\tx}(v)$\phi_{\tx}\tx(v). One classical property of Shapley values is that, given $v(\varnothing) = 0$v(\varnothing) = 0, as is the case for us by \ref{sched:empty-set}, their sum equals the valuation of the \emph{grand coalition} $\txset$\txset: $\sum_{\tx \in \txset}\phi_{\tx}(v) = v(\txset)$\sum_{\tx \in \txset}\tx \in \txset\phi_{\tx}\tx(v) = v(\txset). The proof is straightforward: for any fixed ordering $\sigma$\sigma of $\txset$\txset, the sum of the marginal contributions of the transactions is a telescoping sum, i.e., except first and last terms, all others appear once positively and once negatively, canceling as a result and leaving us with $v(\txset) - v(\varnothing) = v(\txset)$v(\txset) - v(\varnothing) = v(\txset). Since the sum does not depend on $\sigma$\sigma, the same holds when taking the expectation with respect to $\sigma$\sigma. This already gives us our first property of the Shapley \GCM, namely Efficiency:
|
2502.11964
|
Proposals
| 228,272
| 92
|
22,511,878
| 16
|
A second \GCM based on the idea that a transaction's gas consumption should be its marginal contribution to the execution time is the \emph{Banzhaf} mechanism:
|
2502.11964
|
Proposals
| 228,272
| 95
|
22,511,879
| 17
|
\begin{definition}[Banzhaf \GCM]\label{def:banzhaf-gas-comp-mech}
Given a set of transactions $\txset$
consisting of $|\txset| = n$ transactions
and a transaction $\tx \in \txset$, the \emph{Banzhaf} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^B(\tx) := \frac{1}{2^{n - 1}}\sum_{S\subseteq \txset\setminus\{\tx\}} \left[ v ( S\cup\{ \tx\} ) - v(S) \right].
\end{equation*}
\end{definition}\begin{definition}[Banzhaf \GCM]\label{def:banzhaf-gas-comp-mech}
Given a set of transactions $\txset$
consisting of $|\txset| = n$ transactions
and a transaction $\tx \in \txset$, the \emph{Banzhaf} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^B(\tx) := \frac{1}{2^{n - 1}}\sum_{S\subseteq \txset\setminus\{\tx\}} \left[ v ( S\cup\{ \tx\} ) - v(S) \right].
\end{equation*}
\end{definition}\label{def:banzhaf-gas-comp-mech}
Given a set of transactions $\txset$\txset
consisting of $|\txset| = n$|\txset| = n transactions
and a transaction $\tx \in \txset$\tx \in \txset, the \emph{Banzhaf} \GCM computes the amount of gas used by $\tx$\tx as follows:
\gas_{\txset}\txset^B(\tx) := \frac{1}{2^{n - 1}}\sum_{S\subseteq \txset\setminus\{\tx\}}S\subseteq \txset\setminus\{\tx\} \left[ v ( S\cup\{ \tx\} ) - v(S) \right].
\vspace{2pt}
|
2502.11964
|
Proposals
| 228,272
| 96
|
22,511,880
| 18
|
As for the Shapley mechanism, the Banzhaf mechanism can be understood probabilistically, but this time with a separate experiment for each transaction $\tx$\tx. Sample a subset $S$S of transactions other than $\tx$\tx uniformly at random and compute $\tx$\tx's marginal contribution to the execution time when added to $S$S, namely $v(S \cup \{\tx\}) - v(S)$v(S \cup \{\tx\}) - v(S). The gas consumption of $\tx$\tx is then its expected marginal contribution to the execution time. The familiar reader will recognize this as the definition of the \emph{Banzhaf power index} $\beta_{\tx}(v)$\beta_{\tx}\tx(v). For consistency with \emph{Shapley values}, we will instead call these \emph{Banzhaf values}. While more straightforward than the corresponding experiment used in defining Shapley values, the fact that we now have $n$n separate experiments makes the sum of the values no longer well-behaved, losing Efficiency for the Banzhaf mechanism:
|
2502.11964
|
Proposals
| 228,272
| 97
|
22,511,881
| 19
|
\begin{restatable}{lemma}{banzhafnotefficient}\label{lemma:banzhaf-not-efficient} The Banzhaf \GCM does \emph{not} satisfy Efficiency (\cref{prop:efficiency}).
\end{restatable}{lemma}lemma{banzhafnotefficient}banzhafnotefficient\label{lemma:banzhaf-not-efficient} The Banzhaf \GCM does \emph{not} satisfy Efficiency (\cref{prop:efficiency}).
|
2502.11964
|
Proposals
| 228,272
| 98
|
22,511,882
| 20
|
We conclude this section by introducing two additional reasonable mechanisms \emph{without} Easy Gas Estimation. These mechanisms are notably more straightforward than the Shapley and Banzhaf \GCMs, as they avoid computing marginal contributions. However, they have other drawbacks that will become more apparent when we begin studying their normative properties alongside the other mechanisms.
|
2502.11964
|
Proposals
| 228,272
| 99
|
22,511,883
| 21
|
\begin{definition}[Time-Proportional Makespan \GCM]\label{def:time-prop-makespan-gas-comp-mech}
Given a set of transactions $\txset$
and a transaction $\tx \in \txset$, the \emph{Time-Proportional Makespan (TPM)} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^\textit{TPM}(\tx) := \frac{t(\tx)}{\sum_{\tx' \in T}t(\tx')} \cdot v(T).
\end{equation*}
\end{definition}\begin{definition}[Time-Proportional Makespan \GCM]\label{def:time-prop-makespan-gas-comp-mech}
Given a set of transactions $\txset$
and a transaction $\tx \in \txset$, the \emph{Time-Proportional Makespan (TPM)} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^\textit{TPM}(\tx) := \frac{t(\tx)}{\sum_{\tx' \in T}t(\tx')} \cdot v(T).
\end{equation*}
\end{definition}\label{def:time-prop-makespan-gas-comp-mech}
Given a set of transactions $\txset$\txset
and a transaction $\tx \in \txset$\tx \in \txset, the \emph{Time-Proportional Makespan (TPM)} \GCM computes the amount of gas used by $\tx$\tx as follows:
\gas_{\txset}\txset^\textit{TPM}(\tx) := \frac{t(\tx)}{\sum_{\tx' \in T}t(\tx')} \cdot v(T).
\vspace{2pt}
|
2502.11964
|
Proposals
| 228,272
| 100
|
22,511,884
| 22
|
\begin{definition}[Equally-Split Makespan \GCM]\label{def:eq-split-makespan-gas-comp-mech}
Given a set of transactions $\txset$
and a transaction $\tx \in \txset$, the \emph{Equally-Split Makespan (ESM)} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^\textit{ESM}(\tx) := \frac{v(\txset)}{|\txset|}.
\end{equation*}
\end{definition}\begin{definition}[Equally-Split Makespan \GCM]\label{def:eq-split-makespan-gas-comp-mech}
Given a set of transactions $\txset$
and a transaction $\tx \in \txset$, the \emph{Equally-Split Makespan (ESM)} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^\textit{ESM}(\tx) := \frac{v(\txset)}{|\txset|}.
\end{equation*}
\end{definition}\label{def:eq-split-makespan-gas-comp-mech}
Given a set of transactions $\txset$\txset
and a transaction $\tx \in \txset$\tx \in \txset, the \emph{Equally-Split Makespan (ESM)} \GCM computes the amount of gas used by $\tx$\tx as follows:
\gas_{\txset}\txset^\textit{ESM}(\tx) := \frac{v(\txset)}{|\txset|}.
\vspace{2pt}
|
2502.11964
|
Proposals
| 228,272
| 101
|
22,511,886
| 24
|
\begin{definition}[Exponentially-Split Makespan \GCM]\label{def:exp-split-makespan-gas-comp-mech}
Given a set of transactions $\txset$
and a transaction $\tx \in \txset$, the \emph{Exponentially-Split Makespan (XSM)} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^\textit{XSM}(\tx) := \frac{v(\txset)}{3^{|\txset|}}.
\end{equation*}
\end{definition}\begin{definition}[Exponentially-Split Makespan \GCM]\label{def:exp-split-makespan-gas-comp-mech}
Given a set of transactions $\txset$
and a transaction $\tx \in \txset$, the \emph{Exponentially-Split Makespan (XSM)} \GCM computes the amount of gas used by $\tx$ as follows:%
\begin{equation*}
\gas_{\txset}^\textit{XSM}(\tx) := \frac{v(\txset)}{3^{|\txset|}}.
\end{equation*}
\end{definition}\label{def:exp-split-makespan-gas-comp-mech}
Given a set of transactions $\txset$\txset
and a transaction $\tx \in \txset$\tx \in \txset, the \emph{Exponentially-Split Makespan (XSM)} \GCM computes the amount of gas used by $\tx$\tx as follows:
\gas_{\txset}\txset^\textit{XSM}(\tx) := \frac{v(\txset)}{3^{|\txset|}}.
\vspace{2pt}
|
2502.11964
|
Proposals
| 228,272
| 103
|
22,511,888
| 1
|
In this section, we provide a detailed analysis of the normative properties of our \GCMs. Our results are summarized in \cref{tbl:results}.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 105
|
22,511,889
| 2
|
\caption{Comparison of \GCMs based on their adherence to the defined properties. \(\,<\,\) indicates that the mechanism \emph{strictly} satisfies the property,
\(\,=\,\) indicates \emph{trivial} satisfaction (by equality),
and \(\,\leq\,\) indicates satisfaction (not necessarily strict).
A ``\(\checkmark\)'' means the property is satisfied;
``\(\times\)'' means it is not satisfied.
For computational complexity,
``\(v\)'' means the mechanism is as hard to compute as \(v(\cdot)\) itself,
``\(S(v)\)'' means it is as hard as computing Shapley values for \(v\),
and ``\(B(v)\)'' means it is as hard as computing Banzhaf values for \(v\).
}
\label{tbl:results}\vspace{-20pt}
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 106
|
22,511,891
| 4
|
In this section, we analyze the current and Weighted Area \GCMs. By definition, both satisfy Easy Gas Estimation (\cref{prop:Composability}) and are Poly-time Computable (\cref{prop:computational_complexity}). Since they satisfy Easy Gas Estimation but are not constant \GCMs, \cref{thm:impossibility} implies that \emph{neither} satisfies Scheduling Monotonicity (\cref{prop:schedule_monotonicity}) nor Efficiency (\cref{prop:efficiency}). Furthermore, both mechanisms satisfy strict Time Monotonicity (\cref{prop:time_monotonicity}): increasing the time $t$t of a transaction strictly increases its gas consumption. Similarly, both satisfy strict Set Inclusion (\cref{prop:set_inclusion}): since transactions' gas consumptions are strictly positive, a strict superset of a given set of transactions consumes strictly more gas. For the remaining three properties, the two mechanisms behave slightly differently:
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 108
|
22,511,892
| 5
|
\emph{Storage Key Monotonicity (\cref{prop:resource_monotonicity}).} The current mechanism ignores the set of storage keys $K$K that a transaction accesses, so replacing $K$K with a strict superset of it does not change the transaction's gas consumption. Therefore, the current mechanism satisfies Storage Key Monotonicity \emph{with equality}. On the other hand, the Weighted Area mechanism adds an extra term of $t \cdot w_k > 0$t \cdot w_k > 0 to the gas consumption for each additional storage key $k$k added to $K$K, so it satisfies strict Storage Key Monotonicity.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 109
|
22,511,893
| 6
|
\emph{Storage Key-Time Monotonicity (\cref{prop:resource_time_monotonicity}).} From the above results on whether our two mechanisms satisfy Storage Key Monotonicity and Time Monotonicity, \cref{lemma:prop-1-2-3-equivalence,lemma:prop-1-2-3-equivalence-strict} allow us to conclude that the current mechanism satisfies Storage Key-Time Monotonicity, while the Weighted Area mechanism satisfies it strictly.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 110
|
22,511,894
| 7
|
\emph{Transaction Bundling (\cref{prop:tx_bundling}).} Concatenating two transactions with times $t_1$t_1 and $t_2$t_2 results in a transaction with time $t_1 + t_2$t_1 + t_2. Since the current mechanism equates gas consumption with time, the bundled transaction has the same gas consumption as the two individual transactions combined. This implies that the current mechanism satisfies Transaction Bundling \emph{with equality}. Finally, the Weighted Area mechanism satisfies Transaction Bundling, as shown below. If we restrict ourselves to bundling transactions with different storage key sets, the property is strictly satisfied.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 111
|
22,511,895
| 8
|
\begin{lemma} The Weighted Area \GCM satisfies Transaction Bundling (\cref{prop:tx_bundling}). If we only consider bundling transactions with different storage key sets, the property is strictly satisfied.
\end{lemma}\begin{lemma} The Weighted Area \GCM satisfies Transaction Bundling (\cref{prop:tx_bundling}). If we only consider bundling transactions with different storage key sets, the property is strictly satisfied.
\end{lemma} The Weighted Area \GCM satisfies Transaction Bundling (\cref{prop:tx_bundling}). If we only consider bundling transactions with different storage key sets, the property is strictly satisfied.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 112
|
22,511,896
| 9
|
\begin{proof} Consider two transactions $\tx_1 \simeq (t_1, K_1)$ and $\tx_2 \simeq (t_2, K_2)$. Let $\tx_3 \simeq (t_1 + t_2, K_1 \cup K_2)$ be a transaction consisting of the concatenation of $\tx_1$ and $\tx_2$. We want to show that $\gas^\textit{WA}(\tx_1)+\gas^\textit{WA}(\tx_2) \leq \gas^\textit{WA}(\tx_3)$. By definition, this amounts to:
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 113
|
22,511,897
| 10
|
\begin{equation}
t_1 \cdot \left(1 + \sum_{k \in K_1} w_k\right) +
t_2 \cdot \left(1 + \sum_{k \in K_2} w_k\right) \leq
(t_1 + t_2) \cdot \left(1 + \sum_{k \in K_1 \cup K_2} w_k\right)
\end{equation}
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 114
|
22,511,898
| 11
|
Which is true because $\sum_{k \in K_i} w_k \leq \sum_{k \in K_1 \cup K_2} w_k$ for $i \in \{1, 2\}$. Because the weights are strictly positive, equality occurs if and only if $K_1 = K_1 \cup K_2$ and $K_2 = K_1 \cup K_2$; i.e., $K_1 = K_2$. Hence, the property is satisfied strictly if we restrict bundling transactions to cases where $K_1 \neq K_2$.
\end{proof}\begin{proof} Consider two transactions $\tx_1 \simeq (t_1, K_1)$ and $\tx_2 \simeq (t_2, K_2)$. Let $\tx_3 \simeq (t_1 + t_2, K_1 \cup K_2)$ be a transaction consisting of the concatenation of $\tx_1$ and $\tx_2$. We want to show that $\gas^\textit{WA}(\tx_1)+\gas^\textit{WA}(\tx_2) \leq \gas^\textit{WA}(\tx_3)$. By definition, this amounts to:
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 115
|
22,511,899
| 12
|
\begin{equation}
t_1 \cdot \left(1 + \sum_{k \in K_1} w_k\right) +
t_2 \cdot \left(1 + \sum_{k \in K_2} w_k\right) \leq
(t_1 + t_2) \cdot \left(1 + \sum_{k \in K_1 \cup K_2} w_k\right)
\end{equation}
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 116
|
22,511,900
| 13
|
Which is true because $\sum_{k \in K_i} w_k \leq \sum_{k \in K_1 \cup K_2} w_k$ for $i \in \{1, 2\}$. Because the weights are strictly positive, equality occurs if and only if $K_1 = K_1 \cup K_2$ and $K_2 = K_1 \cup K_2$; i.e., $K_1 = K_2$. Hence, the property is satisfied strictly if we restrict bundling transactions to cases where $K_1 \neq K_2$.
\end{proof} Consider two transactions $\tx_1 \simeq (t_1, K_1)$\tx_1 \simeq (t_1, K_1) and $\tx_2 \simeq (t_2, K_2)$\tx_2 \simeq (t_2, K_2). Let $\tx_3 \simeq (t_1 + t_2, K_1 \cup K_2)$\tx_3 \simeq (t_1 + t_2, K_1 \cup K_2) be a transaction consisting of the concatenation of $\tx_1$\tx_1 and $\tx_2$\tx_2. We want to show that $\gas^\textit{WA}(\tx_1)+\gas^\textit{WA}(\tx_2) \leq \gas^\textit{WA}(\tx_3)$\gas^\textit{WA}(\tx_1)+\gas^\textit{WA}(\tx_2) \leq \gas^\textit{WA}(\tx_3). By definition, this amounts to:
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 117
|
22,511,901
| 14
|
t_1 \cdot \left(1 + \sum_{k \in K_1}k \in K_1 w_k\right) +
t_2 \cdot \left(1 + \sum_{k \in K_2}k \in K_2 w_k\right) \leq
(t_1 + t_2) \cdot \left(1 + \sum_{k \in K_1 \cup K_2}k \in K_1 \cup K_2 w_k\right)
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 118
|
22,511,902
| 15
|
Which is true because $\sum_{k \in K_i} w_k \leq \sum_{k \in K_1 \cup K_2} w_k$\sum_{k \in K_i}k \in K_i w_k \leq \sum_{k \in K_1 \cup K_2}k \in K_1 \cup K_2 w_k for $i \in \{1, 2\}$i \in \{1, 2\}. Because the weights are strictly positive, equality occurs if and only if $K_1 = K_1 \cup K_2$K_1 = K_1 \cup K_2 and $K_2 = K_1 \cup K_2$K_2 = K_1 \cup K_2; i.e., $K_1 = K_2$K_1 = K_2. Hence, the property is satisfied strictly if we restrict bundling transactions to cases where $K_1 \neq K_2$K_1 \neq K_2.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 119
|
22,511,903
| 16
|
\subsection{Mechanisms \emph{without} Easy Gas Estimation}
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 120
|
22,511,904
| 17
|
In this section, we analyze the Shapley, Banzhaf, TPM, ESM, and XSM \GCMs. By definition, all of them require knowledge of $T$T to compute $\gas_T(\tx)$\gas_T(\tx), so they do not satisfy Easy Gas Estimation (\cref{prop:Composability}). Next, we examine each of the remaining properties individually and analyze whether they hold for our five mechanisms.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 121
|
22,511,905
| 18
|
\emph{Poly-time Computable (\cref{prop:computational_complexity}).} The TPM, ESM, and XSM \GCMs are all Poly-time Computable whenever determining the makespan of a given set of transactions under the chosen scheduler is feasible in polynomial time, i.e., when $v(T)$v(T) can be computed in polynomial time.\footnote{Notably, this is essentially never true for non-trivial makespan minimization problems. In particular, if we restrict our attention to the case of unit-length transactions (i.e., with $t = 1$) and infinitely many threads $n = \infty$, then checking for the existence of a schedule with makespan $c$ corresponds to checking whether the intersection graph of the transactions (i.e., where edges correspond to transactions with intersecting storage key sets) is $c$-colorable. For $c = 3$, this is well-known to be NP-complete, but this result is for general graphs. However, there is a straightforward way to model any graph $G = (V, E)$ as an intersection graph of transactions: vertices $v \in V$ are transactions and for every edge $(u, v) \in E$, create a new storage key $k$ and add it to the storage key sets of transactions $u$ and $v$.} In fact, the problems are all equally difficult to computing such makespans. In contrast, computing gas consumptions under the Shapley and Banzhaf \GCMs hits another hurdle: it is no longer enough to be able to efficiently compute $v(T)$v(T), but instead, one needs to be able to compute the Shapley or Banzhaf values of the transactions, involving aggregating over $(v(T'))_{T' \subseteq T}$(v(T'))_{T' \subseteq T}T' \subseteq T.
Hence, computing Shapley and Banzhaf values is, in general, NP-hard and \#P-complete \cite{shapley_banzhaf_1,shapley_banzhaf_2,shapley_banzhaf_3,shapley_banzhaf_4,shapley_banzhaf_5}. Moreover, no deterministic polynomial-time algorithm can approximate the Shapley values within a constant factor unless P = NP. Using randomization could, in principle, circumvent this: to approximate an average consisting of exponentially many terms, sample polynomially many uniformly at random, and take their average. One could imagine implementing this in a blockchain via a VRF or similar mechanism to ensure unbiased randomness.
Still, this comes with a notable downside: the accuracy of the computed gas consumption may be hard to predict in advance and might vary wildly.
Moreover, even with randomization, one could still run into issues computing the sampled terms, which is as hard as evaluating $v$v a constant number of times. See \cite{cooperative_book_edith} for an ampler discussion of computing Shapley and Banzhaf values. Note that the previous results mostly pertain to functions $v$v of a different shape than ours (i.e., not related to scheduling). We have not attempted to show that hardness is retained in our context, but expect this to be true. Note still that because the Shapley \GCM is Efficient, it can be used to compute $v(T)$v(T) by adding up the gas consumptions of all transactions in $T$T, meaning that the Shapley \GCM is at least as difficult to compute as $v(T)$v(T). However, this simple reduction no longer works for the Banzhaf \GCM.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 122
|
22,511,906
| 19
|
\emph{Efficiency (\cref{prop:efficiency}).} We have already seen that the Shapley \GCM is Efficient (\cref{lemma:shapley-efficient}) while the Banzhaf \GCM is not (\cref{lemma:banzhaf-not-efficient}). Moreover, by adding up the gas consumption, one can immediately see from the definitions that TPM and ESM are Efficient, while XSM is not. We take this occasion to also note that any \GCM can be made Efficient by appropriately scaling its output. Most notably, applying this to the Banzhaf \GCM yields a mechanism based on the well-known \emph{normalized Banzhaf values}, though the resulting mechanism otherwise seems rather poorly behaved.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 123
|
22,511,907
| 20
|
\emph{Scheduling Monotonicity (\cref{prop:schedule_monotonicity}).} The ESM and XSM \GCMs can be easily seen to strictly satisfy Scheduling Monotonicity. This is because, under both mechanisms, given a set of transactions $T$T, the gas consumption of any transaction in $T$T is computed as $f(|T|) \cdot v(T)$f(|T|) \cdot v(T), where $f$f is either $\frac{1}{x}$\frac{1}{x} or $\frac{1}{3^x}$\frac{1}{3^x}. Notably, the first factor depends only on $|T|$|T|, so if a transaction in $T$T were modified in a way that increases the makespan, this increase would also be reflected proportionally in its gas consumption. In contrast, the Shapley, Banzhaf, and TPM \GCMs do \emph{not} satisfy Scheduling Monotonicity (proven in \cref{lem:shapleyschedule} below). This may be particularly surprising for the Shapley and Banzhaf \GCMs, as they were specifically designed to account for marginal increases in makespan. The catch is that Scheduling Monotonicity considers replacing a transaction $\tx$\tx with another one leading to an increase in makespan. However, some elements in $\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}$\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}S\subseteq \txset\setminus\{\tx\} may still decrease as a result (except for $S = \txset\setminus\{\tx\}$S = \txset\setminus\{\tx\}, for which we assumed an increase). Because both the Shapley and Banzhaf values of $\tx$\tx take a weighted average over these values, the average might still decrease, which is what happens in the example in the lemma below.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 124
|
22,511,908
| 21
|
\begin{lemma}\label{lem:shapleyschedule} The
Shapley, Banzhaf and TPM \GCMs do \emph{not} satisfy Scheduling Monotonicity (\cref{prop:schedule_monotonicity}).
\end{lemma}\begin{lemma}\label{lem:shapleyschedule} The
Shapley, Banzhaf and TPM \GCMs do \emph{not} satisfy Scheduling Monotonicity (\cref{prop:schedule_monotonicity}).
\end{lemma}\label{lem:shapleyschedule} The
Shapley, Banzhaf and TPM \GCMs do \emph{not} satisfy Scheduling Monotonicity (\cref{prop:schedule_monotonicity}).
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 125
|
22,511,909
| 22
|
\begin{proof} Consider the set of transactions \(T=\{\tx_1,\tx_2\}\) and two additional transactions $\tx_3$ and $\tx_4$ such that $\tx_1\simeq(1,\{k_1\})$, $\tx_2\simeq(3,\{k_2\})$, $\tx_3\simeq(2,\{k_1\})$, and $\tx_4\simeq(1,\{k_2\})$. Then, for any number of threads $n \geq 2$ and the optimal scheduler we have $v(\txset \cup \{\tx_3\}) =3 < 4 = v(\txset\cup\{\tx_4\})$. However:
%
\begin{align*}
\gas_{\txset\cup\{\tx_3\}}^S(\tx_3)=\frac{2+2+2+0+0+0}{6} = 1 >& \frac{5}{6} = \frac{1+1+1+1+1+0}{6} = \gas_{\txset\cup\{\tx_4\}}^S(\tx_4)\\
%
\gas_{\txset\cup\{\tx_3\}}^B(\tx_3)=\frac{2+2+0+0}{4} = 1 > & \frac{3}{4} = \frac{1+0+1+1}{4} = \gas_{\txset\cup\{\tx_4\}}^B(\tx_4) \\
%
\gas_{\txset\cup\{\tx_3\}}^\textit{TPM}(\tx_3)=\frac{2}{1+3+2}\cdot 3 = 1 > & \frac{4}{5} = \frac{1}{1+3+1} \cdot 4 = \gas_{\txset\cup\{\tx_4\}}^\textit{TPM}(\tx_4).
\end{align*}
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 126
|
22,511,910
| 23
|
So, the Shapley, Banzhaf, and TPM \GCMs all violate Scheduling Monotonicity in this case.
\end{proof}\begin{proof} Consider the set of transactions \(T=\{\tx_1,\tx_2\}\) and two additional transactions $\tx_3$ and $\tx_4$ such that $\tx_1\simeq(1,\{k_1\})$, $\tx_2\simeq(3,\{k_2\})$, $\tx_3\simeq(2,\{k_1\})$, and $\tx_4\simeq(1,\{k_2\})$. Then, for any number of threads $n \geq 2$ and the optimal scheduler we have $v(\txset \cup \{\tx_3\}) =3 < 4 = v(\txset\cup\{\tx_4\})$. However:
%
\begin{align*}
\gas_{\txset\cup\{\tx_3\}}^S(\tx_3)=\frac{2+2+2+0+0+0}{6} = 1 >& \frac{5}{6} = \frac{1+1+1+1+1+0}{6} = \gas_{\txset\cup\{\tx_4\}}^S(\tx_4)\\
%
\gas_{\txset\cup\{\tx_3\}}^B(\tx_3)=\frac{2+2+0+0}{4} = 1 > & \frac{3}{4} = \frac{1+0+1+1}{4} = \gas_{\txset\cup\{\tx_4\}}^B(\tx_4) \\
%
\gas_{\txset\cup\{\tx_3\}}^\textit{TPM}(\tx_3)=\frac{2}{1+3+2}\cdot 3 = 1 > & \frac{4}{5} = \frac{1}{1+3+1} \cdot 4 = \gas_{\txset\cup\{\tx_4\}}^\textit{TPM}(\tx_4).
\end{align*}
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 127
|
22,511,911
| 24
|
So, the Shapley, Banzhaf, and TPM \GCMs all violate Scheduling Monotonicity in this case.
\end{proof} Consider the set of transactions \(T=\{\tx_1,\tx_2\}\)T=\{\tx_1,\tx_2\} and two additional transactions $\tx_3$\tx_3 and $\tx_4$\tx_4 such that $\tx_1\simeq(1,\{k_1\})$\tx_1\simeq(1,\{k_1\}), $\tx_2\simeq(3,\{k_2\})$\tx_2\simeq(3,\{k_2\}), $\tx_3\simeq(2,\{k_1\})$\tx_3\simeq(2,\{k_1\}), and $\tx_4\simeq(1,\{k_2\})$\tx_4\simeq(1,\{k_2\}). Then, for any number of threads $n \geq 2$n \geq 2 and the optimal scheduler we have $v(\txset \cup \{\tx_3\}) =3 < 4 = v(\txset\cup\{\tx_4\})$v(\txset \cup \{\tx_3\}) =3 < 4 = v(\txset\cup\{\tx_4\}). However:
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 128
|
22,511,912
| 25
|
\gas_{\txset\cup\{\tx_3\}}\txset\cup\{\tx_3\}^S(\tx_3)=\frac{2+2+2+0+0+0}{6} = 1 >& \frac{5}{6} = \frac{1+1+1+1+1+0}{6} = \gas_{\txset\cup\{\tx_4\}}\txset\cup\{\tx_4\}^S(\tx_4)\\
\gas_{\txset\cup\{\tx_3\}}\txset\cup\{\tx_3\}^B(\tx_3)=\frac{2+2+0+0}{4} = 1 > & \frac{3}{4} = \frac{1+0+1+1}{4} = \gas_{\txset\cup\{\tx_4\}}\txset\cup\{\tx_4\}^B(\tx_4) \\
\gas_{\txset\cup\{\tx_3\}}\txset\cup\{\tx_3\}^\textit{TPM}(\tx_3)=\frac{2}{1+3+2}\cdot 3 = 1 > & \frac{4}{5} = \frac{1}{1+3+1} \cdot 4 = \gas_{\txset\cup\{\tx_4\}}\txset\cup\{\tx_4\}^\textit{TPM}(\tx_4).
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 129
|
22,511,914
| 27
|
\emph{Transaction Bundling (\cref{prop:tx_bundling}).} We find that the Banzhaf, TPM and XSM \GCMs satisfy Transaction Bundling, only XSM satisfying it strictly, while the Shapley and ESM \GCMs do not satisfy the property. We prove these facts in the following 5 lemmas.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 131
|
22,511,915
| 28
|
\begin{restatable}{lemma}{lemmashapleynotxbundling} The Shapley \GCM does \emph{not} satisfy Transaction Bundling (\cref{prop:tx_bundling}).
\end{restatable}{lemma}lemma{lemmashapleynotxbundling}lemmashapleynotxbundling The Shapley \GCM does \emph{not} satisfy Transaction Bundling (\cref{prop:tx_bundling}).
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 132
|
22,511,916
| 29
|
\begin{restatable}{lemma}{lemmabanzhafhastxbundl} The Banzhaf \GCM satisfies Transaction Bundling (\cref{prop:tx_bundling}).
\end{restatable}{lemma}lemma{lemmabanzhafhastxbundl}lemmabanzhafhastxbundl The Banzhaf \GCM satisfies Transaction Bundling (\cref{prop:tx_bundling}).
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 133
|
22,511,917
| 30
|
\begin{restatable}{lemma}{lemmatpmhasbundling} The TPM \GCM satisfies Transaction Bundling (\cref{prop:tx_bundling}).
\end{restatable}{lemma}lemma{lemmatpmhasbundling}lemmatpmhasbundling The TPM \GCM satisfies Transaction Bundling (\cref{prop:tx_bundling}).
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 134
|
22,511,918
| 31
|
\begin{restatable}{lemma}{lemmaesmnobundling} The ESM \GCM does \emph{not} satisfy Transaction Bundling (\cref{prop:tx_bundling}).
\end{restatable}{lemma}lemma{lemmaesmnobundling}lemmaesmnobundling The ESM \GCM does \emph{not} satisfy Transaction Bundling (\cref{prop:tx_bundling}).
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 135
|
22,511,926
| 39
|
\emph{Storage Key-Time Monotonicity (\cref{prop:resource_time_monotonicity}).} All five mechanisms satisfy this property. For the ESM and XSM \GCMs, this is an immediate consequence of property \ref{sched:monot-t-R}. For the Shapley and Banzhaf \GCMs, one can see this by recalling that they compute the gas consumption of a transaction $\tx \in T$\tx \in T as a weighted average over $\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}$\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}S\subseteq \txset\setminus\{\tx\}. Hence, by property \ref{sched:monot-t-R}, when $\tx$\tx is replaced with some $\tx' \gtrsim \tx$\tx' \gtrsim \tx, no term in the previous decreases, so their weighted average also does not decrease. Finally, for the TPM \GCM, we show this in the lemma after the next paragraph*.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 143
|
22,511,927
| 40
|
\emph{Storage Key Monotonicity (\cref{prop:resource_monotonicity}) and Time Monotonicity (\cref{prop:time_monotonicity}).} Because all five mechanisms satisfy Storage Key-Time Monotonicity, by \cref{lemma:prop-1-2-3-equivalence}, they also all satisfy Storage Key Monotonicity and Time Monotonicity. Out of the five mechanisms, the Shapley, Banzhaf, and TPM \GCMs satisfy the property strictly. For the first two, this is because when $t(\tx)$t(\tx) increases, at least one term in $\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}$\left(v(S\cup\{ \tx\} ) - v(S)\right)_{S\subseteq \txset\setminus\{\tx\}}S\subseteq \txset\setminus\{\tx\} strictly increases, namely the term for $S = \varnothing$S = \varnothing, while no terms decrease by property \ref{sched:monot-t-R}. Last, for the TPM \GCM, we show this in the lemma below.
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 144
|
22,511,961
| 41
|
\begin{restatable}{lemma}{lemmatpmrestimesensandtimesens} The TPM \GCM satisfies Storage Key-Time Monotonicity (\cref{prop:resource_time_monotonicity}) and strict Time Monotonicity (\cref{prop:time_monotonicity}).
\end{restatable}{lemma}lemma{lemmatpmrestimesensandtimesens}lemmatpmrestimesensandtimesens The TPM \GCM satisfies Storage Key-Time Monotonicity (\cref{prop:resource_time_monotonicity}) and strict Time Monotonicity (\cref{prop:time_monotonicity}).
|
2502.11964
|
Analysis of Our GCMs
| 228,273
| 145
|
22,511,962
| 0
|
\label{sec:outlook}
|
2502.11964
|
Towards a Fee Market for Parallel Execution
| 228,274
| 146
|
22,511,963
| 1
|
Armed with an understanding of the trade-offs between desirable properties in a \GCM, in particular the impossibility of satisfying all properties within a single mechanism, and informed by our analysis of various candidate mechanisms and their properties, we propose two mechanisms for practical implementations of a fee market supporting parallel execution. Each represents one side of the design spectrum: one drawn from the class of mechanisms with Easy Gas Estimation, and the other from the class without it.
|
2502.11964
|
Towards a Fee Market for Parallel Execution
| 228,274
| 147
|
22,511,964
| 2
|
The advantage of adopting a mechanism with Easy Gas Estimation is that satisfying this property ensures that each transaction consumes a fixed amount of gas, regardless of other transactions in the same block. From the perspective of currently deployed TFMs (e.g., EIP‑1559), which process transactions with fixed sizes, nothing changes. Thus, the existing properties of these mechanisms remain intact. On a high level, the key properties we strive for in a TFM are incentive compatibility for both block producers and users, welfare optimality, and collusion resistance. However, no TFM can achieve all these properties simultaneously~\cite{chung2023foundations,chung2024collusion,gafni2024barriers}. Importantly, when composing a \GCM that satisfies Easy Gas Estimation with a TFM of choice, the level of sophistication required from users in their bidding strategy does not increase, unlike in currently deployed fee markets for parallel execution~\cite{lostin2025truth,mueller2025}.
|
2502.11964
|
Towards a Fee Market for Parallel Execution
| 228,274
| 148
|
22,511,965
| 3
|
On the other hand, foregoing Easy Gas Estimation makes it possible to price transactions according to the load they impose on the network relative to the other transactions in the block, rather than evaluating each one in isolation. Mechanisms in this class can align the total gas charged in a block with the actual execution cost of that block, enabling resource pricing at the block level rather than only at the transaction level.
Recent discussions in Ethereum research have explored similar designs under the umbrella of block-level fee markets~\cite{ethresearch_blockfee2025}. If the protocol is already moving toward block-level metering for other resources, such as data availability or storage contention, extending this approach to execution costs could enable direct efficiency gains from parallel execution that are not achievable with transaction-level pricing alone.
|
2502.11964
|
Towards a Fee Market for Parallel Execution
| 228,274
| 149
|
22,511,966
| 4
|
For the class of mechanisms with Easy Gas Estimation, the natural choice is the Weighted Area \GCM. Given that it satisfies Easy Gas Estimation, it achieves the most additional properties one could hope for in a (non-constant) \GCM (see \cref{thm:impossibility}).
|
2502.11964
|
Towards a Fee Market for Parallel Execution
| 228,274
| 150
|
22,511,968
| 5
|
For the class of mechanisms without Easy Gas Estimation, we propose the Time-Proportional Makespan \GCM as a concrete candidate. By allocating gas in proportion to each transaction’s execution time relative to the block’s total makespan, it provides a simple way to align gas consumption with how transactions constrain parallel execution within a block.
|
2502.11964
|
Towards a Fee Market for Parallel Execution
| 228,274
| 151
|
22,511,969
| 6
|
The decision between these approaches ultimately depends on protocol-level priorities: maintaining user-side simplicity and compatibility with existing TFMs through Easy Gas Estimation, which also makes implementation simpler by reusing current transaction-level pricing infrastructure, or pursuing block-level efficiency and more accurate resource pricing at the cost of giving up the previous benefits.
|
2502.11964
|
Towards a Fee Market for Parallel Execution
| 228,274
| 152
|
22,511,970
| 0
|
\subsection{Transaction Fee Mechanisms}
|
2502.11964
|
Related Work
| 228,275
| 153
|
22,511,971
| 1
|
There is extensive research on blockchain fee markets, with a particular focus on Ethereum and Bitcoin. Early studies primarily examined Bitcoin, exploring monopolistic pricing mechanisms~\cite{lavi2022redesigning,yao2020incentive}. More recent contributions to this field include~\cite{nisan2023serial,gafni2022greedy,penna2024serial}. Unlike these works, our study concerns measuring storage key usage on a blockchain with client-side parallel execution, rather than focusing on pricing.
|
2502.11964
|
Related Work
| 228,275
| 154
|
22,511,982
| 12
|
While structurally similar, the blockchain context introduces an additional constraint. In addition to objectives like maximizing provider revenue, it emphasizes maximizing throughput by ensuring transactions are processed efficiently, minimizing their collective execution time within a block. Moreover, blockchain transactions explicitly declare access to distinct storage keys, introducing contention through overlapping access patterns. This adds complexity compared to cloud models, which typically abstract resource demands as scalar quantities and address contention at a more aggregated level.
|
2502.11964
|
Related Work
| 228,275
| 165
|
22,511,972
| 2
|
The TFM design framework was introduced by Roughgarden~\cite{roughgarden2020transaction,roughgarden2024transaction}. Roughgarden's analysis of the EIP-1559 mechanism~\cite{eip1559spec} initiated an active line of research on TFMs. Chung and Shi~\cite{chung2023foundations} demonstrated that no TFM can be ideal --- meaning it cannot simultaneously be incentive-compatible for users and block producers while also being resistant to collusion between the two. This conclusion holds even for weaker definitions of collusion resilience, as shown by Chung et al.~\cite{chung2024collusion} and Gafni and Yaish \cite{gafni2024barriers}. Finally, attempts to address these limitations using cryptographic techniques \cite{shi2022can,wu2023maximizing} have made progress in overcoming certain impossibilities, while other attempts relax the desiderata~\cite{gafni2024discrete}. However, designing an ideal TFM still remains out of reach. While these studies examine the limitations of TFMs, our focus is on GCMs for parallel execution and how to integrate them with a TFM.
|
2502.11964
|
Related Work
| 228,275
| 155
|
22,511,973
| 3
|
A related body of work examines the dynamics of TFMs over multiple blocks, particularly focusing on the base fee in EIP-1559. Leonardos et al.~\cite{leonardos2021dynamical,leonardos2023optimality} demonstrate that the stability of the base fee depends on the adjustment parameter, with short-term volatility but long-term block size stability. Reijsbergen et al.~\cite{reijsbergen2021transaction} suggest using an adaptive adjustment parameter to mitigate block size fluctuations, while Ferreira et al.~\cite{ferreira2021dynamic} highlight user experience issues caused by bounded base fee oscillations. Additionally, Hougaard and Pourpouneh~\cite{hougaard2023farsighted} and Azouvi et al.~\cite{azouvi2023base} reveal that the base fee can be manipulated by non-myopic miners.
|
2502.11964
|
Related Work
| 228,275
| 156
|
22,511,974
| 4
|
Given the discussion surrounding multi-dimensional fees in Ethereum~\cite{multidimensional_eip1559,buterin2024multidim} and the deployment of EIP-4844~\cite{eip4844} (a first step towards a multi-dimensional fee market on Ethereum), a recent line of work explores multi-dimensional fee markets, focusing on efficient pricing mechanisms and their optimality. This work is further refined by Diamandis et al.~\cite{diamandis2023designing}, who design and analyze multi-dimensional blockchain fee markets to align incentives and improve network performance. Building on this, Angeris et al.~\cite{angeris2024multidimensional} prove that such fee markets are nearly optimal, with efficiency improving over time even under adversarial conditions. Multidimensional fee markets are closely related to fee markets designed for parallel execution. In particular, in the weighted area \GCM, the weights can be interpreted as fees within a multidimensional fee market. Unlike previous literature on multidimensional fee markets, we focus on parallelization, introduce desirable properties, and evaluate how various mechanisms perform.
|
2502.11964
|
Related Work
| 228,275
| 157
|
22,511,975
| 5
|
Further extensions of TFMs have emerged. Bahrani et al.~\cite{bahrani2024transaction} consider TFMs in the presence of maximal extractable value (MEV), i.e., value extractable by the block producer. Further, Wang et al.~\cite{wang2024mechanism} design a fee mechanism for proof networks, whereas Bahrani et al.~\cite{bahrani2024resonance} introduce a transaction fee mechanism for heterogeneous computation. Our work most closely relates to the latter, but, in contrast, our chosen approach is closer to multidimensional fee markets, trading complexity for the block producer for stronger incentive compatibility for the user.
|
2502.11964
|
Related Work
| 228,275
| 158
|
22,511,976
| 6
|
Local fee markets have recently been a topic of discussion in the blockchain space~\cite{eclipse2024local,diamandis2024toward,lostin2025truth,keyneom2024local}. The core idea is that transactions interacting with highly contested states incur higher fees, while those involving non-contested states pay lower fees. However, discussions on local fee markets have largely remained high-level, without a precise characterization of the desired properties beyond this general goal. Moreover, currently implemented local fee markets~\cite{lostin2025truth} require significant user sophistication to set fees appropriately. In this work, we formalize the desiderata for fee markets in the context of parallel execution and identify the weighted area \GCM as a promising candidate. One key advantage is its compatibility with a TFM, enabling simple fee estimation for users.
|
2502.11964
|
Related Work
| 228,275
| 159
|
22,511,978
| 8
|
Blockchain concurrency has been a focal point in an active line of research. In particular, numerous efforts have aimed to enable parallel transaction processing through speculative execution~\cite{sergey2017concurrent,zhang2018enabling,amiri2019parblockchain,dickerson2017adding,anjana2022optsmart,gelashvili2022block,chen2021forerunner,saraph2019empirical}. Note that speculative execution is already deployed by multiple blockchains~\cite{aptos,sei_protocol,monad}. Static analysis has also been employed to identify parallelizable transactions, though it cannot completely eliminate inherent dependencies~\cite{pirlea2021practical,murgia2021theory}. Similarly, Neiheiser et al.~\cite{neiheiser2024pythia} demonstrate how parallel execution can assist struggling nodes in catching up. While these works are orthogonal to ours, they highlight the overhead of parallel execution when there is no advance knowledge about a transaction's state accesses.
|
2502.11964
|
Related Work
| 228,275
| 161
|
22,511,979
| 9
|
Further, Saraph and Herlihy \cite{saraph2019empirical} and Heimbach et al.~\cite{heimbach2023defi} have evaluated the parallelization potential of the Ethereum workload. The latter demonstrates that a speedup of approximately fivefold is achievable, assuming state accesses are known in advance. Additionally, Solana~\cite{solana} and Sui~\cite{sui} already perform parallel execution with advance knowledge of state accesses. However, in practice, state accesses are not known beforehand on many blockchains such as Ethereum. There, less than 2\% of transactions disclose them proactively, as shown by Heimbach et al.~\cite{heimbach2023dissecting}, due to a lack of incentives. In this work, we aim to take a step toward unlocking the parallelization potential by designing a TFM that supports parallel execution. This mechanism relies on the disclosure of state accesses as done in Solana~\cite{solana} and Sui~\cite{sui}.
|
2502.11964
|
Related Work
| 228,275
| 162
|
22,511,981
| 11
|
While our primary focus is on transaction fee mechanisms for blockchain networks, parallels can be drawn to resource allocation and pricing in cloud computing. Several papers study pricing schemes for cloud computing resources in a monopolistic setting, where a single provider sells access to multiple resources and sets prices to optimize revenue~\cite{10.1145/2479942.2479944}. In these models, the provider posts prices for resource usage, users select resource bundles to maximize their individual surplus, and the provider earns revenue from the resulting allocation. Mechanisms such as CloudPack~\cite{inproceedings} extend this approach by incorporating workload flexibility, using concepts like Shapley values for fair cost distribution among colocated jobs. Similarly, works such as~\cite{Pei_2024} address resource contention through dynamic pricing strategies that adjust costs based on runtime slowdowns.
|
2502.11964
|
Related Work
| 228,275
| 164
|
22,511,983
| 13
|
Capturing the effects of storage key-level conflicts on parallel execution in blockchains requires fee computation mechanisms that are sensitive to these fine-grained interactions. Thus, while conceptually addressing similar challenges, the blockchain setting imposes unique constraints that motivate the need for specialized GCMs as explored in this work.
|
2502.11964
|
Related Work
| 228,275
| 166
|
22,511,984
| 0
|
In this work, we took a step towards creating a fee market that meets the demands of parallel execution environments while also upholding the properties we want from a TFM.
|
2502.11964
|
Conclusion
| 228,276
| 167
|
22,511,985
| 1
|
Recently, the idea of local fee markets has been proposed for blockchains that support parallel execution. However, to the best of our knowledge, before this work, the demands on these fee markets have only been outlined at a very high level, and the markets that have been implemented are not ideal yet, e.g., they require high levels of sophistication from users when bidding.
|
2502.11964
|
Conclusion
| 228,276
| 168
|
22,511,986
| 2
|
In this work, we addressed this gap by introducing a framework with two key components: a GCM, which measures the execution-related load a transaction imposes on the network in units of gas, and a TFM, which determines the cost associated with each unit of gas. We then formalized the desired properties for the GCM in such a fee market. After outlining these desiderata, we evaluated various mechanisms against them and identified two strong candidates through this analysis: the \textit{weighted area} \GCM for the class of mechanisms with Easy Gas Estimation, and the \textit{time-proportional makespan} \GCM for the class without it.
|
2502.11964
|
Conclusion
| 228,276
| 169
|
22,511,987
| 3
|
Setting the right incentives in fee markets for parallel execution is crucial to unlocking the full potential of execution layer parallelization, and we hope that our work contributes to the development of fee markets capable of meeting the demands of such environments.
|
2502.11964
|
Conclusion
| 228,276
| 170
|
22,511,988
| 4
|
\bibliography{lipics-v2021-sample-article}
|
2502.11964
|
Conclusion
| 228,276
| 171
|
22,511,989
| 0
|
\subsection{Transaction Bundling (\cref{prop:tx_bundling})}
\lemmashapleynotxbundling*
\begin{proof} Consider the set of transactions \(\txset=\{\tx_4\}\) and three other transactions $\tx_1, \tx_2, \tx_3$ such that $\tx_3$ is the concatenation of $\tx_1$ and $\tx_2$, where $\tx_1\simeq(1,\{k_2\})$, $\tx_2\simeq(1,\{k_2\})$, $\tx_3\simeq(2,\{k_2\})$, and $\tx_4\simeq(1,\{k_1\})$. Then, for any number of threads $n \geq 2$ and the optimal scheduler, we have:
%
\begin{align*}
\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_1)=\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_2)&=\frac{0+1+1+1+1+1}{6} \\
%
\gas_{\txset \cup \{\tx_3\}}^S(\tx_3)&=\frac{2+1}{2}.
\end{align*}
|
2502.11964
|
Proofs Omitted From
| 228,277
| 172
|
22,511,990
| 1
|
Thus, \(\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_1)+\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_2)=\frac{10}{6}>\frac{3}{2}= \gas_{\txset \cup \{\tx_3\}}^S(\tx_3)\), which violates Transaction Bundling.
\end{proof}\begin{proof} Consider the set of transactions \(\txset=\{\tx_4\}\) and three other transactions $\tx_1, \tx_2, \tx_3$ such that $\tx_3$ is the concatenation of $\tx_1$ and $\tx_2$, where $\tx_1\simeq(1,\{k_2\})$, $\tx_2\simeq(1,\{k_2\})$, $\tx_3\simeq(2,\{k_2\})$, and $\tx_4\simeq(1,\{k_1\})$. Then, for any number of threads $n \geq 2$ and the optimal scheduler, we have:
%
\begin{align*}
\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_1)=\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_2)&=\frac{0+1+1+1+1+1}{6} \\
%
\gas_{\txset \cup \{\tx_3\}}^S(\tx_3)&=\frac{2+1}{2}.
\end{align*}
|
2502.11964
|
Proofs Omitted From
| 228,277
| 173
|
22,511,991
| 2
|
Thus, \(\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_1)+\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_2)=\frac{10}{6}>\frac{3}{2}= \gas_{\txset \cup \{\tx_3\}}^S(\tx_3)\), which violates Transaction Bundling.
\end{proof} Consider the set of transactions \(\txset=\{\tx_4\}\)\txset=\{\tx_4\} and three other transactions $\tx_1, \tx_2, \tx_3$\tx_1, \tx_2, \tx_3 such that $\tx_3$\tx_3 is the concatenation of $\tx_1$\tx_1 and $\tx_2$\tx_2, where $\tx_1\simeq(1,\{k_2\})$\tx_1\simeq(1,\{k_2\}), $\tx_2\simeq(1,\{k_2\})$\tx_2\simeq(1,\{k_2\}), $\tx_3\simeq(2,\{k_2\})$\tx_3\simeq(2,\{k_2\}), and $\tx_4\simeq(1,\{k_1\})$\tx_4\simeq(1,\{k_1\}). Then, for any number of threads $n \geq 2$n \geq 2 and the optimal scheduler, we have:
|
2502.11964
|
Proofs Omitted From
| 228,277
| 174
|
22,511,992
| 3
|
\gas_{\txset \cup \{\tx_1, \tx_2\}}\txset \cup \{\tx_1, \tx_2\}^S(\tx_1)=\gas_{\txset \cup \{\tx_1, \tx_2\}}\txset \cup \{\tx_1, \tx_2\}^S(\tx_2)&=\frac{0+1+1+1+1+1}{6} \\
\gas_{\txset \cup \{\tx_3\}}\txset \cup \{\tx_3\}^S(\tx_3)&=\frac{2+1}{2}.
|
2502.11964
|
Proofs Omitted From
| 228,277
| 175
|
22,511,993
| 4
|
Thus, \(\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_1)+\gas_{\txset \cup \{\tx_1, \tx_2\}}^S(\tx_2)=\frac{10}{6}>\frac{3}{2}= \gas_{\txset \cup \{\tx_3\}}^S(\tx_3)\)\gas_{\txset \cup \{\tx_1, \tx_2\}}\txset \cup \{\tx_1, \tx_2\}^S(\tx_1)+\gas_{\txset \cup \{\tx_1, \tx_2\}}\txset \cup \{\tx_1, \tx_2\}^S(\tx_2)=\frac{10}{6}>\frac{3}{2}= \gas_{\txset \cup \{\tx_3\}}\txset \cup \{\tx_3\}^S(\tx_3), which violates Transaction Bundling.
|
2502.11964
|
Proofs Omitted From
| 228,277
| 176
|
22,511,994
| 5
|
\lemmabanzhafhastxbundl*
\begin{proof}
Consider a set of transactions $\txset$ and three transactions $\tx_1, \tx_2, \tx_3 \notin T$ such that $\tx_3$ is the concatenation of $\tx_1$ and $\tx_2$. Then, for any scheduler satisfying property \ref{sched:monot-bundle}, we have:
\begin{align*}
\gas^B_{\txset\cup \{\tx_1, \tx_2\}}(\tx_1) &+ \gas^B_{\txset\cup \{\tx_1, \tx_2\}}(\tx_2)
\\&= \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_1\}} \left[ v ( S\cup\{ \tx_1\} ) - v(S) \right] \\&+ \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_2\}} \left[ v ( S\cup\{ \tx_2\} ) - v(S) \right]
\\&=\frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset} \left[ v ( S\cup\{ \tx_1\} ) - v(S) + v ( S\cup\{ \tx_1, \tx_2\} ) - v(S\cup\{\tx_2\}) \right] \\&+ \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset} \left[ v ( S\cup\{ \tx_2\} ) - v(S) + v ( S\cup\{ \tx_1,\tx_2\} ) - v(S\cup\{\tx_1\}) \right]
\\&=\frac{2}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset} \left[v ( S\cup\{ \tx_1, \tx_2\} ) - v(S)\right]
\\&\leq\frac{1}{2^{|\txset\cup \{\tx_3\}| - 1}}\sum_{S\subseteq \txset} \left[v ( S\cup\{ \tx_3\} ) - v(S)\right] \quad = \quad \gas^B_{\txset\cup \{\tx_3\}}(\tx_3). \qedhere
\end{align*}
\end{proof}\begin{proof}
Consider a set of transactions $\txset$ and three transactions $\tx_1, \tx_2, \tx_3 \notin T$ such that $\tx_3$ is the concatenation of $\tx_1$ and $\tx_2$. Then, for any scheduler satisfying property \ref{sched:monot-bundle}, we have:
\begin{align*}
\gas^B_{\txset\cup \{\tx_1, \tx_2\}}(\tx_1) &+ \gas^B_{\txset\cup \{\tx_1, \tx_2\}}(\tx_2)
\\&= \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_1\}} \left[ v ( S\cup\{ \tx_1\} ) - v(S) \right] \\&+ \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_2\}} \left[ v ( S\cup\{ \tx_2\} ) - v(S) \right]
\\&=\frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset} \left[ v ( S\cup\{ \tx_1\} ) - v(S) + v ( S\cup\{ \tx_1, \tx_2\} ) - v(S\cup\{\tx_2\}) \right] \\&+ \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset} \left[ v ( S\cup\{ \tx_2\} ) - v(S) + v ( S\cup\{ \tx_1,\tx_2\} ) - v(S\cup\{\tx_1\}) \right]
\\&=\frac{2}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset} \left[v ( S\cup\{ \tx_1, \tx_2\} ) - v(S)\right]
\\&\leq\frac{1}{2^{|\txset\cup \{\tx_3\}| - 1}}\sum_{S\subseteq \txset} \left[v ( S\cup\{ \tx_3\} ) - v(S)\right] \quad = \quad \gas^B_{\txset\cup \{\tx_3\}}(\tx_3). \qedhere
\end{align*}
\end{proof}
Consider a set of transactions $\txset$\txset and three transactions $\tx_1, \tx_2, \tx_3 \notin T$\tx_1, \tx_2, \tx_3 \notin T such that $\tx_3$\tx_3 is the concatenation of $\tx_1$\tx_1 and $\tx_2$\tx_2. Then, for any scheduler satisfying property \ref{sched:monot-bundle}, we have:
\gas^B_{\txset\cup \{\tx_1, \tx_2\}}\txset\cup \{\tx_1, \tx_2\}(\tx_1) &+ \gas^B_{\txset\cup \{\tx_1, \tx_2\}}\txset\cup \{\tx_1, \tx_2\}(\tx_2)
\\&= \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_1\}}S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_1\} \left[ v ( S\cup\{ \tx_1\} ) - v(S) \right] \\&+ \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_2\}}S\subseteq (\txset\cup\{\tx_1, \tx_2\})\setminus\{\tx_2\} \left[ v ( S\cup\{ \tx_2\} ) - v(S) \right]
\\&=\frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset}S\subseteq \txset \left[ v ( S\cup\{ \tx_1\} ) - v(S) + v ( S\cup\{ \tx_1, \tx_2\} ) - v(S\cup\{\tx_2\}) \right] \\&+ \frac{1}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset}S\subseteq \txset \left[ v ( S\cup\{ \tx_2\} ) - v(S) + v ( S\cup\{ \tx_1,\tx_2\} ) - v(S\cup\{\tx_1\}) \right]
\\&=\frac{2}{2^{|\txset\cup \{\tx_1, \tx_2\}| - 1}}\sum_{S\subseteq \txset}S\subseteq \txset \left[v ( S\cup\{ \tx_1, \tx_2\} ) - v(S)\right]
\\&\leq\frac{1}{2^{|\txset\cup \{\tx_3\}| - 1}}\sum_{S\subseteq \txset}S\subseteq \txset \left[v ( S\cup\{ \tx_3\} ) - v(S)\right] \quad = \quad \gas^B_{\txset\cup \{\tx_3\}}\txset\cup \{\tx_3\}(\tx_3). \qedhere
|
2502.11964
|
Proofs Omitted From
| 228,277
| 177
|
22,512,026
| 6
|
\lemmatpmhasbundling*
\begin{proof}
Consider a set of transactions $\txset$ and three transactions $\tx_1, \tx_2, \tx_3 \notin T$ such that $\tx_3$ is the concatenation of $\tx_1$ and $\tx_2$.
Then, for any scheduler satisfying property \ref{sched:monot-bundle}, we have:
\begin{align*}
\gas^\textit{TPM}_{\txset\cup \{\tx_1, \tx_2\}}(\tx_1) &+ \gas^\textit{TPM}_{\txset\cup \{\tx_1, \tx_2\}}(\tx_2)
\\&=\frac{t(\tx_1)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})+\frac{t(\tx_2)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&=\frac{t(\tx_1)+t(\tx_2)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&=\frac{t(\tx_3)}{\sum_{\tx' \in \txset\cup\{\tx_3\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&\leq\frac{t(\tx_3)}{\sum_{\tx' \in \txset\cup\{\tx_3\}}t(\tx')} \cdot v(\txset\cup\{\tx_3\}) \quad = \quad \gas^\textit{TPM}_{\txset\cup \{\tx_3\}}(\tx_3). \qedhere
\end{align*}
\end{proof}\begin{proof}
Consider a set of transactions $\txset$ and three transactions $\tx_1, \tx_2, \tx_3 \notin T$ such that $\tx_3$ is the concatenation of $\tx_1$ and $\tx_2$.
Then, for any scheduler satisfying property \ref{sched:monot-bundle}, we have:
\begin{align*}
\gas^\textit{TPM}_{\txset\cup \{\tx_1, \tx_2\}}(\tx_1) &+ \gas^\textit{TPM}_{\txset\cup \{\tx_1, \tx_2\}}(\tx_2)
\\&=\frac{t(\tx_1)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})+\frac{t(\tx_2)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&=\frac{t(\tx_1)+t(\tx_2)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&=\frac{t(\tx_3)}{\sum_{\tx' \in \txset\cup\{\tx_3\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&\leq\frac{t(\tx_3)}{\sum_{\tx' \in \txset\cup\{\tx_3\}}t(\tx')} \cdot v(\txset\cup\{\tx_3\}) \quad = \quad \gas^\textit{TPM}_{\txset\cup \{\tx_3\}}(\tx_3). \qedhere
\end{align*}
\end{proof}
Consider a set of transactions $\txset$\txset and three transactions $\tx_1, \tx_2, \tx_3 \notin T$\tx_1, \tx_2, \tx_3 \notin T such that $\tx_3$\tx_3 is the concatenation of $\tx_1$\tx_1 and $\tx_2$\tx_2.
Then, for any scheduler satisfying property \ref{sched:monot-bundle}, we have:
\gas^\textit{TPM}_{\txset\cup \{\tx_1, \tx_2\}}\txset\cup \{\tx_1, \tx_2\}(\tx_1) &+ \gas^\textit{TPM}_{\txset\cup \{\tx_1, \tx_2\}}\txset\cup \{\tx_1, \tx_2\}(\tx_2)
\\&=\frac{t(\tx_1)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})+\frac{t(\tx_2)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&=\frac{t(\tx_1)+t(\tx_2)}{\sum_{\tx' \in \txset\cup\{\tx_1,\tx_2\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&=\frac{t(\tx_3)}{\sum_{\tx' \in \txset\cup\{\tx_3\}}t(\tx')} \cdot v(\txset\cup\{\tx_1,\tx_2\})
\\&\leq\frac{t(\tx_3)}{\sum_{\tx' \in \txset\cup\{\tx_3\}}t(\tx')} \cdot v(\txset\cup\{\tx_3\}) \quad = \quad \gas^\textit{TPM}_{\txset\cup \{\tx_3\}}\txset\cup \{\tx_3\}(\tx_3). \qedhere
|
2502.11964
|
Proofs Omitted From
| 228,277
| 178
|
22,512,087
| 7
|
\lemmaesmnobundling*
\begin{proof} Consider the set of transactions \(\txset=\{\tx_4\}\) and three transactions $\tx_1, \tx_2, \tx_3 \notin \txset$ such that $\tx_3$ is the concatenation of $\tx_1$ and $\tx_2$, where $\tx_1\simeq(1,\{k_1\})$, $\tx_2\simeq(1,\{k_1\})$, $\tx_3\simeq(2,\{k_1\})$, and $\tx_4\simeq(1,\{k_1\})$. Then, for any number of threads $n \geq 2$ and the optimal scheduler, we have:
\[
\gas_{\txset \cup \{\tx_1, \tx_2\}}^\textit{ESM}(\tx_1)=\gas_{\txset \cup \{\tx_1, \tx_2\}}^\textit{ESM}(\tx_2) = \frac{3}{3}, \quad \gas_{\txset \cup \{\tx_3\}}^\textit{ESM}(\tx_3) = \frac{3}{2}.
\]
|
2502.11964
|
Proofs Omitted From
| 228,277
| 179
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.