Hound: a parallel image distribution system for cluster based on Docker

doi:10.23919/JSEE.2023.000105

Journal of Systems Engineering and Electronics ›› 2023, Vol. 34 ›› Issue (4): 955-965.doi: 10.23919/JSEE.2023.000105

• SYSTEMS ENGINEERING • Previous Articles

Hound: a parallel image distribution system for cluster based on Docker

Zijie LIU¹^,²(), Junjiang LI¹^,²(), Can CHEN²^,³(), Dengyin ZHANG¹^,²^,*()

¹ School of Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
² Jiangsu Key Laboratory of Broadband Wireless Communication and Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
³ College of Telecommunications & Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China

Received:2021-10-15 Online:2023-08-18 Published:2023-08-28
Contact: Dengyin ZHANG E-mail:2019070274@njupt.edu.cn;admin@run-linux.com;chencan@njupt.edu.cn;zhangdy@njupt.edu.cn
About author:
LIU Zijie was born in 1996. He received his B.S. degree from Nanjing Institute of Technology, Nanjing, China, in 2014. He is currently pursuing his Ph.D. degree in information network with Nanjing University of Posts and Telecommunications, Nanjing, China. His current research interests include cloud computing and distributed systems. E-mail: 2019070274@njupt.edu.cn

LI Junjiang was born in 1996. He received his B.S. degree from Tongda College of Nanjing University of Posts and Telecommunications, Yangzhou, China, in 2014. He is currently pursuing his M.S. degree in logistics engineering with Nanjing University of Posts and Telecommunications, Nanjing, China. His current research interests include cloud computing, container orchestration systems, and distributed systems. E-mail: admin@run-linux.com

CHEN Can was born in 1993. He received his B.S. degree from Nanjing University of Posts and Telecommunications, Nanjing, China, in 2015. He is currently pursuing his Ph.D. degree in signal and information processing in the College of Telecommunications and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing, China. His research interests include image and video coding, image and video processing, machine learning, and compressive sensing. E-mail: chencan@njupt.edu.cn

ZHANG Dengyin was born in 1964. He received his B.S., M.S. and Ph.D. degrees from Nanjing University of Posts and Telecommunications, Nanjing, China, in 1986, 1989 and 2004 respectively. He was in Digital Media Lab at Umea University in Sweden as a visiting scholar from 2007 to 2008. He is currently a professor with the School of Internet of Things, Nanjing University of Posts and Telecommunications, Nanjing, China. His research interests include signal and information processing, networking technique, and information security. E-mail: zhangdy@njupt.edu.cn
Supported by:
This work was supported by the National Natural Science Foundation of China (61872423), Industry Prospective Primary Research & Development Plan of Jiangsu Province (BE2017111), the Scientific Research Foundation of the Higher Education Institutions of Jiangsu Province (19KJA180006), and the Postgraduate Research & Practice Innovation Program of Jiangsu Province (KYCX20_0764)

Abstract

Abstract:

Current applications, consisting of multiple replicas, are packaged into lightweight containers with their execution dependencies. Considering the dominant impact of distribution efficiency of gigantic images on container startup (e.g., distributed deep learning application), the image “warm-up” technique which prefetches images of these replicas to destination nodes in the cluster is proposed. However, the current image “warm-up” technique solely focuses on identical image distribution, which fails to take effect when distributing different images to destination nodes. To address this problem, this paper proposes Hound, a simple but efficient cluster image distribution system based on Docker. To support diverse image distribution requests of cluster nodes, Hound additionally adopts node-level parallelism (i.e., downloading images to destination nodes in parallel) to further improve the efficiency of image distribution. The experimental results demonstrate Hound outperforms Docker, kubernetes container runtime interface (CRI-O), and Docker-compose in terms of image distribution performance when cluster nodes request different images. Moreover, the high scalability of Hound is evaluated in the scenario of ten nodes.

Key words: container image, image distribution, parallelism, containerization

Zijie LIU, Junjiang LI, Can CHEN, Dengyin ZHANG. Hound: a parallel image distribution system for cluster based on Docker[J]. Journal of Systems Engineering and Electronics, 2023, 34(4): 955-965.

Figures/Tables 14

Fig 1

Table 1

Table 2

Fig 2

Table 3

Fig 3

Table 4

Table 5

Fig 4

Fig 5

Table 6

Fig 6

Table 7

Table 8

References 25

1	MERKEL D. Docker: lightweight linux containers for consistent development and deployment. Linux Journal, 2014. https://www.linuxjournal.com/content/docker-lightweight-linux-containers-consistent-development-and-deployment.
2	SMET P, DHOEDT B, SIMOENS P Docker layer placement for on-demand provisioning of services on edge clouds. IEEE Trans. on Network and Service Management, 2018, 15 (3): 1161- 1174. doi: 10.1109/TNSM.2018.2844187
3	SINGH S, SINGH N. Containers & Docker: emerging roles & future of Cloud technology. Proc. of the 2nd International Conference on Applied and Theoretical Computing and Communication Technology, 2016: 804−807.
4	MERELLI I, FORNARI F, TORDINI F, et al Exploiting Docker containers over Grid computing for a comprehensive study of chromatin conformation in different cell types. Journal of Parallel and Distributed Computing, 2019, 134, 116- 127. doi: 10.1016/j.jpdc.2019.08.002
5	PAHL C Containerization and the paas cloud. IEEE Cloud Computing, 2015, 2 (3): 24- 31. doi: 10.1109/MCC.2015.51
6	THALHEIM J, BHATOTIA P, FONSECA P, et al. CNTR: lightweight {OS} containers. Proc. of the {USENIX} Annual Technical Conference ({USENIX}{ATC} 18), 2018: 199−212.
7	PENG Y H, BAO Y X, CHEN Y R, et al Dl2: a deep learning-driven scheduler for deep learning clusters. IEEE Trans. on Parallel and Distributed Systems, 2021, 32 (8): 1947- 1960. doi: 10.1109/TPDS.2021.3052895
8	WU Y D, MA K H, YAN X, et al Elastic deep learning in multi-tenant GPU clusters. IEEE Trans. on Parallel and Distributed Systems, 2021, 33 (1): 144- 158.
9	ZHOU Q H, GUO S, QU Z H, et al Petrel: heterogeneity-aware distributed deep learning via hybrid synchronization. IEEE Trans. on Parallel and Distributed Systems, 2020, 32 (5): 1030- 1043.
10	ZHAO N N, TARASOV V, ALBAHAR H, et al Large-scale analysis of docker images and performance implications for container storage systems. IEEE Trans. on Parallel and Distributed Systems, 2020, 32 (4): 918- 930.
11	LEE C, KIM S, KIM E A deduplication-enabled P2P protocol for VM image distribution. IEICE Transactions on Information and Systems, 2015, 98 (5): 1108- 1111.
12	DU L, WO T Y, YANG R Y, et al. Cider: a rapid docker container deployment system through sharing network storage. Proc. of the IEEE 19th International Conference on High Performance Computing and Communications, IEEE 15th International Conference on Smart City, IEEE 3rd International Conference on Data Science and Systems, 2017: 332−339.
13	ZHANG S Q, WU S, FAN H, et al. BED: a block-level deduplication-based container deployment framework. Proc. of the International Conference on Green, Pervasive, and Cloud Computing, 2020: 504−518.
14	ZHAO X, ZHANG Y, WU Y W, et al Liquid: a scalable deduplication file system for virtual machine images. IEEE Trans. on Parallel and Distributed Systems, 2013, 25 (5): 1257- 1266.
15	SAHARAN S, SOMANI G, GUPTA G, et al QuickDedup: efficient VM deduplication in cloud computing environments. Journal of Parallel and Distributed Computing, 2020, 139, 18- 31. doi: 10.1016/j.jpdc.2020.01.002
16	HARTER T, SALMON B, LIU R, et al. Slacker: fast distribution with lazy docker containers. Proc. of the 14th {USENIX} Conference on File and Storage Technologies, 2016: 181−195.
17	CIVOLANI L, PIERRE G, BELLAVISTA P. FogDocker: start container now, fetch image later. Proc. of the 12th IEEE/ACM International Conference on Utility and Cloud Computing, 2019: 51−59.
18	ZHENG C, RUPPRECHT L, TARASOV V, et al. Wharf: sharing docker images in a distributed file system. Proc. of the ACM Symposium on Cloud Computing, 2018: 174−185.
19	CHEN J L, LIAQAT D, GABEL M, et al. Poster: an accelerator for fast container-based applications deployment on the edge. Proc. of the IEEE/ACM Symposium on Edge Computing, 2020: 175−177.
20	AHMED A, PIERRE G. Docker container deployment in fog computing infrastructures. Proc. of the IEEE International Conference on Edge Computing, 2018: 1−8.
21	PENG C Y, KIM M, ZHANG Z, et al. VDN: virtual machine image distribution network for cloud data centers. 2012 Proceedings IEEE INFOCOM, 2012: 181−189.
22	ZHANG Z N, LI Z Y, WU K, et al VMThunder: fast provisioning of large-scale virtual machine clusters. IEEE Trans. on Parallel and Distributed Systems, 2014, 25 (12): 3328- 3338. doi: 10.1109/TPDS.2014.7
23	LIANG M Y, SHEN S Q, LI D S, et al. HDID: an efficient hybrid docker image distribution system for datacenters. Proc. of the National Software Application Conference, 2016: 179−194.
24	WANG K J, YANG Y, LI Y, et al. Fid: a faster image distribution system for docker platform. Proc. of the IEEE 2nd International Workshops on Foundations and Applications of Self* Systems, 2017: 191−198.
25	NATHAN S, GHOSH R, MUKHERJEE T, et al. Comicon: a co-operative management system for docker container images. Proc. of the IEEE International Conference on Cloud Engineering, 2017: 116−126.

Node name	Requested image
Node 1	Image A, Image B, Image C
Node 2	Image B, Image D
Node 3	Image A, Image D
Node 4	Image B, Image C, Image D
Node 5	Image A, Image B, Image C, Image D

Node group (request image)	Node included	Node label
Group 1 (Image A)	Node 1, Node 3, Node 5	ImageName: Image A
Group 2 (Image B)	Node 1, Node 2, Node 4, Node 5	ImageName: Image B
Group 3 (Image C)	Node 1, Node 4, Node 5	ImageName: Image C
Group 4 (Image D)	Node 2, Node 3, Node 4, Node 5	ImageName: Image D

Label information	Node information	Token information
LabelName	NodeName	TokenContent
NodeList	KernelVersion	GenerateTime
−	OperatingSystem	UpdateTime
−	DockerVersion	−

Content	Image size
Tomcat	140 MB, 329 MB
Jetty	225 MB
Django	267 MB
Gcc	409 MB
Rust	543 MB
Ros	768 MB
Silverpeas	845 MB
Nuxeo	910 MB
MXnet	1.34 GB
Tensorflow	1.52 GB, 2.25 GB
Pytorch	3.39 GB, 4.2 GB

Request number	Image combination
1	140 MB + 225 MB
2	267 MB + 329 MB
3	409 MB + 543 MB
4	768 MB + 845 MB
5	910 MB + 1.34 GB
6	1.52 GB + 3.39 GB
7	2.25 GB + 4.2 GB
8	140 MB + 267 MB + 409 MB
9	768 MB + 845 MB + 910 MB
10	1.34 GB + 1.52 GB + 3.39 GB

Hound: a parallel image distribution system for cluster based on Docker

RichHTML

PDF (PC)

Knowledge

Abstract

Cite this article

Share this article

Figures/Tables 14

References 25

Related Articles 1

Recommended Articles

Metrics

Comments

Request number	1	2	3	4	5	6	7	8	9	10
Case A	0	3	0	0	3	0	0	0	0	4
Case B	2	0	0	2	0	2	0	0	2	2
Case C	1	1	1	1	1	1	1	1	1	1

Scenario	Resource
Scenario	Average load(idle)	Average load(busy)	CPU usage/%	Disk write rate/(MB/s)	Network bandwidth/(MB/s)
Single worker	0.14	3.98	339.8	63.47	90.98
Five workers	0.14	3.81	342.4	61.83	90.68
Ten workers	0.14	3.72	327.2	60.58	90.27

Scenario	Master response time/ms	Worker response time/μs
Single worker	12.68	183.87
Five workers	17.37	245.24
Ten workers	21.05	268.18