Improving Energy Efﬁciency of Permissioned Blockchains Using FPGAs Nathania Santoso Haris Javaid

2025-05-08 0 0 871.28KB 8 页 10玖币

侵权投诉

Improving Energy Efﬁciency of Permissioned

Blockchains Using FPGAs

Nathania Santoso Haris Javaid

AMD, Singapore

{nathania.santoso, haris.javaid}@amd.com

Abstract—Permissioned blockchains like Hyperledger Fabric

have become quite popular for implementation of enterprise

applications. Recent research has mainly focused on improving

performance of permissioned blockchains without any considera-

tion of their power/energy consumption. In this paper, we conduct

a comprehensive empirical study to understand energy efﬁciency

(throughput/energy) of validator peer in Hyperledger Fabric (a

major bottleneck node). We pick a number of optimizations

for validator peer from literature (allocated CPUs, software

block cache and FPGA based accelerator). First, we propose

a methodology to measure power/energy consumption of the two

resulting compute platforms (CPU-only and CPU+FPGA). Then,

we use our methodology to evaluate energy efﬁciency of a diverse

set of validator peer conﬁgurations, and present many useful

insights. With careful selection of software optimizations and

FPGA accelerator conﬁguration, we improved energy efﬁciency

of validator peer by 10×compared to vanilla validator peer (i.e.,

energy-aware provisioning of validator peer can deliver 10×more

throughput while consuming the same amount of energy). In

absolute terms, this means 23,000 tx/s with power consumption of

118W from a validator peer using software block cache running

on a 4-core server with AMD/Xilinx Alveo U250 FPGA card.

Index Terms—Energy-efﬁcient blockchains, Hyperledger Fab-

ric, FPGA accelerators

I. INTRODUCTION

Blockchain technology is on the rise due to its capability

of executing transactions (which contain business logic in the

form of smart contracts) and storing them in a decentralized

manner (same ledger distributed among multiple nodes). The

distributed ledger contains blocks where each block has a hash

value of itself and the previous block, assuring immutability

of the ledger data. Hence, a network of nodes that implement

a blockchain essentially provides an implementation of a

distributed ledger for applications that will be developed and

deployed on top of that blockchain. There are two types

of blockchains: public (permissionless) and private (permis-

sioned) blockchains. In a public blockchain, nodes do not

require identity authorization to participate in the network. On

the other hand, the identity of a node must be authenticated

cryptographically to execute transactions in a permissioned

blockchain. Bitcoin and Ethereum are examples of public

blockchains, while Hyperledger Fabric [1] is an example of

permissioned blockchain.

Hyperledger Fabric is one of the most popular permissioned

blockchains for enterprise applications [2]. Most of the recent

research on Fabric has only focused on improving its through-

put (transactions per second, shortened as tx/s), overlooking its

power and energy consumption. Sedlmeir et al. [3] estimated

that although a Fabric network consumed 8 orders of magni-

tude less energy than a public blockchain, it still consumed

10×more energy than a centralized system. Furthermore,

Fabric nodes are often deployed in datacenters where energy

consumption has become an issue due to environmental and

climate change concerns. Therefore, there is a need to explore

energy efﬁciency of Fabric nodes. Previous works [4]–[6] have

shown that validator peer node (which runs the validation

of blocks and transactions before committing them to the

ledger) is one of the major bottlenecks. It is a computationally

intensive operation, and thus utilizes signiﬁcant resources

which will result in high power and energy consumption in

the Fabric network.

In this paper, we propose a power/energy measurement

methodology to conduct a comprehensive evaluation of Hyper-

ledger Fabric validator peer in terms of its power/energy con-

sumption and throughput. Typically, Fabric peer is deployed

on a multicore server (i.e., CPU based system). Recently,

Fabric peer has also been shown to run on a multi-core server

with hardware accelerator on an FPGA card (i.e., CPU+FPGA

based system) for high throughput [6]. In this context, this

paper has the following contributions:

•We propose a power/energy measurement methodology

for a CPU based system, where our approach handles

varying number of vCPUs allocated from the physical

cores of a multi-core server. We use this setup to evaluate

energy efﬁciency of vanilla validator peer and a promi-

nent software optimization (block cache from [4]).

•We integrate a power measurement module inside the

hardware accelerator from [6] for power/energy measure-

ment of a CPU+FPGA based system. We use this setup

to evaluate energy efﬁciency of hardware accelerated

validator peer with and without the software block cache.

•We ﬁnally present a comprehensive study of the interac-

tions between power, energy and throughput of validator

peer, and deduce many insights from comparison of

CPU-only and CPU+FPGA systems for energy-aware

provisioning of validator peers in a Fabric network.

Our experiments with Hyperledger Fabric v2.2 LTS running

on a multi-core server with Alveo U250 card show that FPGA

based hardware accelerator combined with software block

cache can deliver 10×more throughput than a CPU-only

system with same energy consumption (153 vs. 15 tx/s/J).

II. BACKGROUND AND PRELIMINARIES

A. Hyperledger Fabric

Hyperledger Fabric is an open-source and enterprise-grade

implementation of a permissioned blockchain. It is not as-

sociated with any cryptocurrency, and has applications in

arXiv:2210.11839v1 [cs.DC] 21 Oct 2022

commit

block verify

tx verify & vscc

statedb read

& mvcc

validator peerorderersendorsing peersclient

idle

Tthreads

tx5

ledger write

statedb write

historydb write

Fig. 1: Transaction ﬂow in Hyperledger Fabric. txN = transaction N, eN =

endorsement for txN, bN = block N, and cN = conﬁrmation for txN. E =

endorsement, O = ordering, and V = validation phases.

many diverse domains such as supply chain, banking/ﬁnance,

healthcare, etc. Figure 1 depicts how a transaction ﬂows

through various nodes of a Fabric network (e.g., endorsing

peers, validator peers, etc. shown on the top).

A client invokes a transaction by sending it to endorsing

peers. Each endorsing peer will simulate the transaction by

computing its input and output (read and write sets) against the

locally stored state database1. Afterwards, the endorsing peer

will send the simulation results along with its digital signature

back to the client. Once the client has enough endorsements,

it will send the transaction along with the endorsements to the

orderer for inclusion into a block. The orderer will create a

new block and broadcast it to all the peers for validation and

subsequent commit to the ledger. Note that each transaction

contains the digital signature of the client that invoked it, while

each block contains the digital signature of the orderer that

created it. In Fabric, all digital signatures are based on Elliptic

Curve Digital Signature Algorithm (ECDSA) scheme.

The validation and commit phase itself consists of several

operations as shown on the right hand side of Figure 1. When

a validator peer receives a block, it will ﬁrst check the syntax

of the block and verify the orderer’s signature on the block

(step 1). Then, the validator peer will check the syntax of

each transaction in the block and verify the client’s signature

on each transaction (tx verify in step 2). Afterwards, for each

transaction in the block, the signatures of endorsing peers from

the transaction’s endorsements will be veriﬁed and evaluated

against an endorsement policy2(validation system chaincode,

shortened as tx vscc in step 2). If the endorsement policy is

satisﬁed, then the transaction is marked as valid.

In step 3, multi-version concurrency control (shortened as

mvcc) checks are applied to mark a transaction as valid/invalid.

Once the entire block has been validated in steps 1–3, in

step 4, the validator peer will commit the block to its ledger,

1The state database contains the current snapshot of the ledger. For example,

for a banking application, it would contain the current value of each account.

2An endorsement policy is associated with a smart contract/chaincode, and

governs the business logic for approval of the transaction; e.g., policy of a

money transfer chaincode between two banks may require valid endorsements

from each bank, i.e., endorsements from Bank1 AND Bank2.

SERVER

FPGA Card

System Conﬁguration

CMAC Subsystem

User Logic Box

@322 MHz

User Logic Box

@250 MHz

Packet Adapter

QDMA Subsystem

AXI-Lite 125MHz

AXI-Stream

322 MHz

Ethernet Port

CPU

PCIe

AXI-Stream

250 MHz

Fig. 2: Overview of OpenNIC shell. The Blockchain Machine hardware is

part of User Logic Box @ 250MHz.

and update its state database (by applying write sets of valid

transactions) and history database (for book keeping). Note

that validator peers only validate and commit incoming blocks,

while endorsing peers do the same in addition to endorsing

incoming transactions.

The vanilla Fabric validator peer incorporates many soft-

ware optimizations for improved throughput. For example, it

uses multiple threads to verify and validate multiple transac-

tions of a block in parallel (taking advantage of multi-core

servers), which is depicted as T threads in Figure 1. Another

software optimization is to use block cache which caches

unmarshaled contents of a block for subsequent accesses, and

has been reported to improve throughput signiﬁcantly (2.33×

in [4] and 1.67×in [5]). However, block cache is not yet part

of the ofﬁcial Fabric codebase, so we implemented it ourselves

in Fabric v2.2 LTS to evaluate its energy efﬁciency.

B. Blockchain Machine

Javaid et al. [6] proposed a hardware accelerator called

Blockchain Machine (BMac) for improving validator peer’s

throughput, instead of relying on software optimizations. The

BMac peer is designed for a multi-core server with a network-

attached FPGA card (connected to the CPU via PCIe bus),

where the CPU runs modiﬁed Fabric software while FPGA

card is programmed with the hardware accelerator. Since this is

the only hardware accelerator proposed for Hyperledger Fabric

so far, we use it as the CPU+FPGA system in this paper.

Figure 2 provides a simpliﬁed overview of the open-

sourced OpenNIC shell [7], which is the basis for BMac

hardware. The OpenNIC shell is an FPGA based NIC shell for

AMD/Xilinx FPGA cards and provides network connectivity

through CMAC/Ethernet port interfacing and CPU connectiv-

ity through QDMA/PCIe interfacing. Consequently, a user-

deﬁned accelerator can be implemented inside the user logic

box where it can access incoming data from the network

through CMAC/Ethernet port while the CPU can access the

output of hardware accelerator through PCIe/QDMA.

The BMac hardware is implemented as user logic box @

250MHz, as shown in Figure 3. The blocks are received in

FPGA card through the CMAC interface. The ﬁrst module,

protocol processor, processes the incoming Ethernet packets

and extracts relevant data, such as block id, transaction ids,

ECDSA signatures, etc. The second module, block processor,

uses this data to validate the block and its transactions,

commits all valid transactions, and then writes the validation

results in a register map (for CPU access through QDMA).

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

ImprovingEnergyEfciencyofPermissionedBlockchainsUsingFPGAsNathaniaSantosoHarisJavaidAMD,Singaporefnathania.santoso,haris.javaidg@amd.comAbstractPermissionedblockchainslikeHyperledgerFabrichavebecomequitepopularforimplementationofenterpriseapplications.Recentresearchhasmainlyfocusedonimprovingperfo...

展开>> 收起<<

Improving Energy Efﬁciency of Permissioned Blockchains Using FPGAs Nathania Santoso Haris Javaid.pdf

共8页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Improving Energy Efﬁciency of Permissioned Blockchains Using FPGAs Nathania Santoso Haris Javaid

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: