POSTER: Improving Datacenter Efficiency Through Partitioning-Aware Scheduling

Harshad Kasture, Xu Ji, Nosayba El-Sayed, Nathan Beckmann, Xiaosong Ma, Daniel Sanchez

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Citation (Scopus)

Abstract

Datacenter servers often colocate multiple applications to improve utilization and efficiency. However, colocated applications interfere in shared resources, e.g., the last-level cache (LLC) and DRAM bandwidth, causing performance inefficiencies. Prior work has proposed two disjoint approaches to address interference. First, techniques that partition shared resources like the LLC can provide isolation and trade performance among colocated applications within a single node. But partitioning techniques are limited by the fixed resource demands of the applications running on the node. Second, interference-aware schedulers try to find resource-compatible applications and schedule them across nodes to improve performance. But prior schedulers are hampered by the lack of partitioning hardware in conventional multicores, and are forced to take conservative colocation decisions, leaving significant performance on the table. We show that memory-system partitioning and scheduling are complementary, and performing them in a coordinated fashion yields significant benefits. We present Shepherd, a joint scheduler and resource partitioner that seeks to maximize cluster-wide throughput. Shepherd uses detailed application profiling data to partition the shared LLC and to estimate the impact of DRAM bandwidth contention among colocated applications. Shepherd's scheduler leverages this information to colocate applications with complementary resource requirements, improving resource utilization and cluster throughput. We evaluate Shepherd in simulation and on a real cluster with hardware support for cache partitioning. When managing mixes of server and scientific applications, Shepherd improves cluster throughput over an unpartitioned system by 38% on average.

Original languageEnglish
Title of host publicationProceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages134-135
Number of pages2
Volume2017-September
ISBN (Electronic)9781467395243
DOIs
Publication statusPublished - 31 Oct 2017
Event26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017 - Portland, United States
Duration: 9 Sep 201713 Sep 2017

Other

Other26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017
CountryUnited States
CityPortland
Period9/9/1713/9/17

Fingerprint

Partitioning
Scheduling
Resources
Scheduler
Cache
Throughput
Dynamic random access storage
Servers
Server
Vertex of a graph
Interference
Bandwidth
Partition
Hardware
Contention
Profiling
Leverage
Isolation
Table
Disjoint

Keywords

  • Cache Partitioning
  • Colocation
  • Datacenters

ASJC Scopus subject areas

  • Software
  • Theoretical Computer Science
  • Hardware and Architecture

Cite this

Kasture, H., Ji, X., El-Sayed, N., Beckmann, N., Ma, X., & Sanchez, D. (2017). POSTER: Improving Datacenter Efficiency Through Partitioning-Aware Scheduling. In Proceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017 (Vol. 2017-September, pp. 134-135). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/PACT.2017.43

POSTER : Improving Datacenter Efficiency Through Partitioning-Aware Scheduling. / Kasture, Harshad; Ji, Xu; El-Sayed, Nosayba; Beckmann, Nathan; Ma, Xiaosong; Sanchez, Daniel.

Proceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017. Vol. 2017-September Institute of Electrical and Electronics Engineers Inc., 2017. p. 134-135.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Kasture, H, Ji, X, El-Sayed, N, Beckmann, N, Ma, X & Sanchez, D 2017, POSTER: Improving Datacenter Efficiency Through Partitioning-Aware Scheduling. in Proceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017. vol. 2017-September, Institute of Electrical and Electronics Engineers Inc., pp. 134-135, 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017, Portland, United States, 9/9/17. https://doi.org/10.1109/PACT.2017.43
Kasture H, Ji X, El-Sayed N, Beckmann N, Ma X, Sanchez D. POSTER: Improving Datacenter Efficiency Through Partitioning-Aware Scheduling. In Proceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017. Vol. 2017-September. Institute of Electrical and Electronics Engineers Inc. 2017. p. 134-135 https://doi.org/10.1109/PACT.2017.43
Kasture, Harshad ; Ji, Xu ; El-Sayed, Nosayba ; Beckmann, Nathan ; Ma, Xiaosong ; Sanchez, Daniel. / POSTER : Improving Datacenter Efficiency Through Partitioning-Aware Scheduling. Proceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017. Vol. 2017-September Institute of Electrical and Electronics Engineers Inc., 2017. pp. 134-135
@inproceedings{e9b53bfb76f0499e986c96760d6d0641,
title = "POSTER: Improving Datacenter Efficiency Through Partitioning-Aware Scheduling",
abstract = "Datacenter servers often colocate multiple applications to improve utilization and efficiency. However, colocated applications interfere in shared resources, e.g., the last-level cache (LLC) and DRAM bandwidth, causing performance inefficiencies. Prior work has proposed two disjoint approaches to address interference. First, techniques that partition shared resources like the LLC can provide isolation and trade performance among colocated applications within a single node. But partitioning techniques are limited by the fixed resource demands of the applications running on the node. Second, interference-aware schedulers try to find resource-compatible applications and schedule them across nodes to improve performance. But prior schedulers are hampered by the lack of partitioning hardware in conventional multicores, and are forced to take conservative colocation decisions, leaving significant performance on the table. We show that memory-system partitioning and scheduling are complementary, and performing them in a coordinated fashion yields significant benefits. We present Shepherd, a joint scheduler and resource partitioner that seeks to maximize cluster-wide throughput. Shepherd uses detailed application profiling data to partition the shared LLC and to estimate the impact of DRAM bandwidth contention among colocated applications. Shepherd's scheduler leverages this information to colocate applications with complementary resource requirements, improving resource utilization and cluster throughput. We evaluate Shepherd in simulation and on a real cluster with hardware support for cache partitioning. When managing mixes of server and scientific applications, Shepherd improves cluster throughput over an unpartitioned system by 38{\%} on average.",
keywords = "Cache Partitioning, Colocation, Datacenters",
author = "Harshad Kasture and Xu Ji and Nosayba El-Sayed and Nathan Beckmann and Xiaosong Ma and Daniel Sanchez",
year = "2017",
month = "10",
day = "31",
doi = "10.1109/PACT.2017.43",
language = "English",
volume = "2017-September",
pages = "134--135",
booktitle = "Proceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017",
publisher = "Institute of Electrical and Electronics Engineers Inc.",

}

TY - GEN

T1 - POSTER

T2 - Improving Datacenter Efficiency Through Partitioning-Aware Scheduling

AU - Kasture, Harshad

AU - Ji, Xu

AU - El-Sayed, Nosayba

AU - Beckmann, Nathan

AU - Ma, Xiaosong

AU - Sanchez, Daniel

PY - 2017/10/31

Y1 - 2017/10/31

N2 - Datacenter servers often colocate multiple applications to improve utilization and efficiency. However, colocated applications interfere in shared resources, e.g., the last-level cache (LLC) and DRAM bandwidth, causing performance inefficiencies. Prior work has proposed two disjoint approaches to address interference. First, techniques that partition shared resources like the LLC can provide isolation and trade performance among colocated applications within a single node. But partitioning techniques are limited by the fixed resource demands of the applications running on the node. Second, interference-aware schedulers try to find resource-compatible applications and schedule them across nodes to improve performance. But prior schedulers are hampered by the lack of partitioning hardware in conventional multicores, and are forced to take conservative colocation decisions, leaving significant performance on the table. We show that memory-system partitioning and scheduling are complementary, and performing them in a coordinated fashion yields significant benefits. We present Shepherd, a joint scheduler and resource partitioner that seeks to maximize cluster-wide throughput. Shepherd uses detailed application profiling data to partition the shared LLC and to estimate the impact of DRAM bandwidth contention among colocated applications. Shepherd's scheduler leverages this information to colocate applications with complementary resource requirements, improving resource utilization and cluster throughput. We evaluate Shepherd in simulation and on a real cluster with hardware support for cache partitioning. When managing mixes of server and scientific applications, Shepherd improves cluster throughput over an unpartitioned system by 38% on average.

AB - Datacenter servers often colocate multiple applications to improve utilization and efficiency. However, colocated applications interfere in shared resources, e.g., the last-level cache (LLC) and DRAM bandwidth, causing performance inefficiencies. Prior work has proposed two disjoint approaches to address interference. First, techniques that partition shared resources like the LLC can provide isolation and trade performance among colocated applications within a single node. But partitioning techniques are limited by the fixed resource demands of the applications running on the node. Second, interference-aware schedulers try to find resource-compatible applications and schedule them across nodes to improve performance. But prior schedulers are hampered by the lack of partitioning hardware in conventional multicores, and are forced to take conservative colocation decisions, leaving significant performance on the table. We show that memory-system partitioning and scheduling are complementary, and performing them in a coordinated fashion yields significant benefits. We present Shepherd, a joint scheduler and resource partitioner that seeks to maximize cluster-wide throughput. Shepherd uses detailed application profiling data to partition the shared LLC and to estimate the impact of DRAM bandwidth contention among colocated applications. Shepherd's scheduler leverages this information to colocate applications with complementary resource requirements, improving resource utilization and cluster throughput. We evaluate Shepherd in simulation and on a real cluster with hardware support for cache partitioning. When managing mixes of server and scientific applications, Shepherd improves cluster throughput over an unpartitioned system by 38% on average.

KW - Cache Partitioning

KW - Colocation

KW - Datacenters

UR - http://www.scopus.com/inward/record.url?scp=85043576939&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85043576939&partnerID=8YFLogxK

U2 - 10.1109/PACT.2017.43

DO - 10.1109/PACT.2017.43

M3 - Conference contribution

AN - SCOPUS:85043576939

VL - 2017-September

SP - 134

EP - 135

BT - Proceedings - 26th International Conference on Parallel Architectures and Compilation Techniques, PACT 2017

PB - Institute of Electrical and Electronics Engineers Inc.

ER -