EdgeKV Decentralized, scalable, and consistent storage for the edge_PingCAP.pdf

tinge

148

13页

2次

2023-09-26

免费下载

Journal of Parallel and Distributed Computing 144 (2020) 28–40

Contents lists available at ScienceDirect

J. Parallel Distrib. Comput.

journal homepage: www.elsevier.com/locate/jpdc

EdgeKV: Decentralized, scalable, and consistent storage for the edge

Karim Sonbol

∗

, Öznur Özkasap

, Ibrahim Al-Oqily

b,c

, Moayad Aloqaily

Department of Computer Engineering, Koç University, Istanbul, Turkey

The Hashemite University, Zarqa, Jordan

Al-Yamamah University, Riyadh, Saudi Arabia

Al Ain University, United Arab Emirates

a r t i c l e i n f o

Article history:

Received 11 March 2020

Received in revised form 29 April 2020

Accepted 19 May 2020

Available online 30 May 2020

Keywords:

Edge computing

Distributed systems

Key–value store

DHT

Consistency

a b s t r a c t

Edge computing moves the computation closer to the data and the data closer to the user to overcome

the high latency communication of cloud computing. Storage at the edge allows data access with high

speeds that enable latency-sensitive applications in areas such as autonomous driving and smart grid.

However, several distributed services are typically designed for the cloud and building an efficient

edge-enabled storage system is challenging because of the distributed and heterogeneous nature of

the edge and its limited resources. In this paper, we propose EdgeKV, a decentralized storage system

designed for the network edge. EdgeKV offers fast and reliable storage, utilizing data replication with

strong consistency guarantees. With a location-transparent and interface-based design, EdgeKV can

scale with a heterogeneous system of edge nodes. We implement a prototype of the EdgeKV modules

in Golang and evaluate it in both the edge and cloud settings on the Grid’5000 testbed. We utilize

the Yahoo! Cloud Serving Benchmark (YCSB) to analyze the system’s performance under realistic

workloads. Our evaluation results show that EdgeKV outperforms the cloud storage setting with both

local and global data access with an average write response time and throughput improvements of

26% and 19% respectively under the same settings. Our evaluations also show that EdgeKV can scale

with the number of clients, without sacrificing performance. Finally, we discuss the energy efficiency

improvement when utilizing edge resources with EdgeKV instead of a centralized cloud.

1. Introduction

Several distributed services need to store their state and

change it or retrieve it at a later time. This state is generally

required to be stored in a reliable, secure, private, efficient, and

cost-effective way. The cloud has been traditionally used for

storing the state of many programs. The cloud provides flexible

pay-what-you-use cost policy, scalability, high security standards,

and high reliability. The cloud is able to provide such features by

hosting clusters of commodity servers in one physically secure

location; namely a data center [5].

Data centers are sparse because of their special requirements

and high costs for building and maintenance. This implies that the

majority of cloud users would be distant from data centers. Such

distances are important because they directly affect the latency of

data transfer between clients and servers, which becomes more

critical in latency-sensitive systems such as autonomous vehicles,

smart grids, and online multiplayer games. Another problem with

∗

Corresponding author.

E-mail addresses: ksonbol16@ku.edu.tr (K. Sonbol), oozkasap@ku.edu.tr

(Ö. Özkasap), ialoqily@ieee.org (I. Al-Oqily), maloqaily@xanalytics.ca

(M. Aloqaily).

cloud storage is data privacy since sending data to the cloud

typically means sharing it with a third-party. Hence, applications

storing sensitive information may need to find an alternative to

the cloud.

To overcome the high latency of the cloud, fields such as edge,

fog, and mist computing came into existence [4,12]. In this paper,

we refer to these entities collectively as edge nodes. These fields

aim to move the data and computation closer to the consumer.

Therefore, instead of doing most of the computation and stor-

age at the cloud, new computation and storage entities can be

introduced between the client and data center in the client-to-

cloud continuum to partially handle these tasks. This not only

reduces the response latency and saves network bandwidth but

also alleviates some of the workload from the cloud. In addition,

storing and processing data at the edge allow utilizing contextual

information to improve data locality and decision making [18].

Moreover, privacy-oriented applications can store their data on

the local edge, or use the cloud for storing only less-sensitive or

aggregate information.

Edge nodes generally have limited computation, storage, net-

work, or power resources. They may have heterogeneous hard-

ware and software architectures, and are not typically located

https://doi.org/10.1016/j.jpdc.2020.05.009

K. Sonbol, Ö. Özkasap, I. Al-Oqily et al. / Journal of Parallel and Distributed Computing 144 (2020) 28–40 29

Table 1

Summary of existing edge-enabled storage solutions.

Project Use case Data content Consistency Fault-tolerance

FogStore [14] Situation-awareness, applications Contextual data Context-based Yes

Vision-Edge [25] Computer vision Feature vectors, key-frames Data-type-based No

FBase [15] Data-intensive fog applications Application data Eventual

Yes

Configuration data Strong

EC+ [33] MMOG Game events Event-type-based No

Workers-KV [9] Web services Web pages Eventual Yes

Fog05 [11] IaaS Server states Eventual No

OpenStack-Edge [20] IaaS Server states Eventual Yes

Dqlite [7] Embedded devices Sensor data Strong Yes

EdgeKV General-purpose Key–Value pairs Strong Yes

in one location. Therefore, they may communicate over a Wide-

Area Network (WAN) and some even use a wireless protocol,

e.g., WiFi or 4G/5G, to communicate together or with clients.

Consequently, designing an edge-enabled storage system is not

trivial and directly using existing cloud-optimized systems in the

edge may not be possible. Instead, either a middleware layer can

be introduced to make the existing software edge-aware or novel

designs can be developed that are specific to the edge or that are

generic enough to work with both cloud and edge.

A few existing works try to build an edge-capable storage

system but they are either domain-specific [21,25], provide only

weak forms of consistency [11], or require a high level of do-

main knowledge and customization to work efficiently [14]. In

this paper, we propose EdgeKV, a decentralized, general-purpose,

scalable, and reliable storage system for the edge. EdgeKV offers

low-latency access, strong consistency notions, high availability,

and minimal customization requirement. Specifically, we make

the following contributions:

• Propose EdgeKV, a novel storage system architecture for the

edge, explain its modules and their interactions, and the

algorithms used.

• Provide fault-tolerance and reliability through replication in

the edge with strong consistency guarantees. Support load-

balancing with a highly-scalable overlay that has minimal

overhead.

• Support two levels of data locality with different latency

guarantees based on application requirements. The separa-

tion of local and global data allows deploying EdgeKV in

different use cases.

• Develop a prototype [28] allowing for heterogeneity in the

system with a hierarchical design and interface abstrac-

tions. Perform comprehensive experimental analysis on dis-

tributed testbed and comparisons with the centralized cloud

storage.

• Present the performance analysis results of EdgeKV, show-

ing its superior performance to the centralized cloud solu-

tion, especially with local data, and evaluate the scalability

of the system with the number of clients and requests. In

addition, we provide a discussion about the energy effi-

ciency aspect of the system. Moreover, we discuss possible

optimizations for scalability, future research directions for

EdgeKV, and useful insights for edge-enabled application

designers.

The rest of the paper is organized as follows. We discuss the

related work in Section 2. Section 3 introduces the design and

system architecture. Section 4 discusses motivating use cases for

EdgeKV. In Section 5, we discuss implementation details and in

Section 6 we present evaluation results. Finally, we conclude

in Section 8 with a summary of the contributions and future

directions.

2. Related work

There exists a variety of work on distributed storage for the

cloud. This ranges from relational databases such as MySQL and

PostgreSQL to NoSQL databases such as Cassandra and Mon-

goDB and includes key–value stores such as TiKV. However, less

work exists for utilizing the edge/fog resources. Nonetheless, we

discuss the relevant existing works next, also summarized in

Table 1.

FogStore [14] is designed for situation-awareness applications

that use data annotated with context information such as loca-

tion or timestamps. FogStore presents a geo-replicated key–value

storage providing differential consistency guarantees based on

the context. While this allows for usage in different scenarios, it

requires the system user to have expert knowledge of the domain

to define a mapping between data and client contexts and the

required consistency level for different queries. Similarly, Vision-

Edge [25] is an application-specific key–value storage solution

for machine vision applications such as smart surveillance cam-

eras. There are two types of data in such applications: latency-

critical feature vectors and key-frames stored for bookkeeping

purposes. FBase [15] is a replication service for data-intensive fog

applications. It provides programmers with a declarative way to

choose replication paths and data flows across geo-distributed

sites, based on user-provided configuration data. FBase guaran-

tees only eventual consistency for the application data and strong

consistency for configuration data. Some works do not handle

fault-tolerance such as Vision-Edge and Edge-Cloud+ (EC+) [33],

while others provide fault-tolerance but with weak forms of

consistency such as Fog05 and Workers-KV [9]. EC+ is an ar-

chitecture augmented by edge computing for Massively Mul-

tiplayer Online Games with Virtual Reality (VR-MMOG). EC+

utilizes the edge for latency-sensitive local view change updates

and leaves global game state updates, with less strict latency

requirements, to the central cloud. CloudFlare’s Workers-KV [9]

utilizes CloudFlare’s global edge network to build low-latency

globally-available key–value storage. Workers-KV only provides

eventual consistency and is mainly useful for building faster and

customized web applications

We note that each work has a specific use case or class of

use cases for which they are designed. For example, Fog05 [11]

and [20] are Infrastructure-as-a-Service (IaaS) frameworks that

are used for managing both cloud and edge resources. Eclipse

fog05 [11] is a virtualization solution for cloud, edge, and fog

resources suitable for heterogeneous systems. It can integrate any

key–value store and provide a location-transparent and unified

view to it from anywhere in the network through a unified

interface, but provides only eventual consistency. OpenStack-

Edge modifies the OpenStack IaaS framework to enable managing

edge resources. It achieves this by replacing its centralized SQL

database with a distributed Redis cluster. Dqlite [7] provides a

distributed, highly available, and lightweight SQLite implementa-

tion suitable for embedded devices. While Dqlite provides strong

of 13

免费下载

tidb

关注

评论