Fastpass: A Centralized “Zero-Queue” Datacenter Network
An ideal datacenter network should provide several properties, including low median and tail latency, high utilization (throughput), fair allocation of network resources between users or applications, deadline-aware scheduling, and congestion (loss) avoidance. Current datacenter networks inherit the...
Main Authors: | , , , , |
---|---|
Other Authors: | |
Format: | Article |
Language: | en_US |
Published: |
Association for Computing Machinery
2014
|
Online Access: | http://hdl.handle.net/1721.1/88141 https://orcid.org/0000-0001-6590-8392 https://orcid.org/0000-0002-4566-771X https://orcid.org/0000-0003-0737-3259 https://orcid.org/0000-0002-1455-9652 |
_version_ | 1826198582822174720 |
---|---|
author | Perry, Jonathan Ousterhout, Amy Elizabeth Balakrishnan, Hari Shah, Devavrat Fugal, Hans |
author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Perry, Jonathan Ousterhout, Amy Elizabeth Balakrishnan, Hari Shah, Devavrat Fugal, Hans |
author_sort | Perry, Jonathan |
collection | MIT |
description | An ideal datacenter network should provide several properties, including low median and tail latency, high utilization (throughput), fair allocation of network resources between users or applications, deadline-aware scheduling, and congestion (loss) avoidance. Current datacenter networks inherit the principles that went into the design of the Internet, where packet transmission and path selection decisions are distributed among the endpoints and routers. Instead, we propose that each sender should delegate control—to a centralized arbiter—of when each packet should be transmitted and what path it should follow. This paper describes Fastpass, a datacenter network architecture built using this principle. Fastpass incorporates two fast algorithms: the first determines the time at which each packet should be transmitted, while the second determines the path to use for that packet. In addition, Fastpass uses an efficient protocol between the endpoints and the arbiter and an arbiter replication strategy for fault-tolerant failover. We deployed and evaluated Fastpass in a portion of Facebook’s datacenter network. Our results show that Fastpass achieves high throughput comparable to current networks at a 240 reduction is queue lengths (4.35 Mbytes reducing to 18 Kbytes), achieves much fairer and consistent flow throughputs than the baseline TCP (5200 reduction in the standard deviation of per-flow throughput with five concurrent connections), scalability from 1 to 8 cores in the arbiter implementation with the ability to schedule 2.21 Terabits/s of traffic in software on eight cores, and a 2.5 reduction in the number of TCP retransmissions in a latency-sensitive service at Facebook. |
first_indexed | 2024-09-23T11:07:07Z |
format | Article |
id | mit-1721.1/88141 |
institution | Massachusetts Institute of Technology |
language | en_US |
last_indexed | 2024-09-23T11:07:07Z |
publishDate | 2014 |
publisher | Association for Computing Machinery |
record_format | dspace |
spelling | mit-1721.1/881412022-09-27T17:14:39Z Fastpass: A Centralized “Zero-Queue” Datacenter Network Perry, Jonathan Ousterhout, Amy Elizabeth Balakrishnan, Hari Shah, Devavrat Fugal, Hans Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Balakrishnan, Hari Perry, Jonathan Ousterhout, Amy Elizabeth Balakrishnan, Hari Shah, Devavrat Fugal, Hans An ideal datacenter network should provide several properties, including low median and tail latency, high utilization (throughput), fair allocation of network resources between users or applications, deadline-aware scheduling, and congestion (loss) avoidance. Current datacenter networks inherit the principles that went into the design of the Internet, where packet transmission and path selection decisions are distributed among the endpoints and routers. Instead, we propose that each sender should delegate control—to a centralized arbiter—of when each packet should be transmitted and what path it should follow. This paper describes Fastpass, a datacenter network architecture built using this principle. Fastpass incorporates two fast algorithms: the first determines the time at which each packet should be transmitted, while the second determines the path to use for that packet. In addition, Fastpass uses an efficient protocol between the endpoints and the arbiter and an arbiter replication strategy for fault-tolerant failover. We deployed and evaluated Fastpass in a portion of Facebook’s datacenter network. Our results show that Fastpass achieves high throughput comparable to current networks at a 240 reduction is queue lengths (4.35 Mbytes reducing to 18 Kbytes), achieves much fairer and consistent flow throughputs than the baseline TCP (5200 reduction in the standard deviation of per-flow throughput with five concurrent connections), scalability from 1 to 8 cores in the arbiter implementation with the ability to schedule 2.21 Terabits/s of traffic in software on eight cores, and a 2.5 reduction in the number of TCP retransmissions in a latency-sensitive service at Facebook. National Science Foundation (U.S.) (grant IIS-1065219) Irwin Mark Jacobs and Joan Klein Jacobs Presidential Fellowship Hertz Foundation (Fellowship) 2014-06-30T17:43:10Z 2014-06-30T17:43:10Z 2014-08 Article http://purl.org/eprint/type/ConferencePaper 978-1-4503-2836-4 http://hdl.handle.net/1721.1/88141 Perry, Jonathan, Amy Ousterhout, Hari Balakrishnan, Devavrat Shah, and Hans Fugal. "Fastpass: A Centralized “Zero-Queue” Datacenter Network." ACM SIGCOMM 2014, Chicago, Illinois, August 17-22, 2014, pp.307-318. https://orcid.org/0000-0001-6590-8392 https://orcid.org/0000-0002-4566-771X https://orcid.org/0000-0003-0737-3259 https://orcid.org/0000-0002-1455-9652 en_US http//dx.doi.org/10.1145/2619239.2626309 SIGCOMM 2014 proceedings Creative Commons Attribution-Noncommercial-Share Alike http://creativecommons.org/licenses/by-nc-sa/4.0/ application/pdf Association for Computing Machinery Perry |
spellingShingle | Perry, Jonathan Ousterhout, Amy Elizabeth Balakrishnan, Hari Shah, Devavrat Fugal, Hans Fastpass: A Centralized “Zero-Queue” Datacenter Network |
title | Fastpass: A Centralized “Zero-Queue” Datacenter Network |
title_full | Fastpass: A Centralized “Zero-Queue” Datacenter Network |
title_fullStr | Fastpass: A Centralized “Zero-Queue” Datacenter Network |
title_full_unstemmed | Fastpass: A Centralized “Zero-Queue” Datacenter Network |
title_short | Fastpass: A Centralized “Zero-Queue” Datacenter Network |
title_sort | fastpass a centralized zero queue datacenter network |
url | http://hdl.handle.net/1721.1/88141 https://orcid.org/0000-0001-6590-8392 https://orcid.org/0000-0002-4566-771X https://orcid.org/0000-0003-0737-3259 https://orcid.org/0000-0002-1455-9652 |
work_keys_str_mv | AT perryjonathan fastpassacentralizedzeroqueuedatacenternetwork AT ousterhoutamyelizabeth fastpassacentralizedzeroqueuedatacenternetwork AT balakrishnanhari fastpassacentralizedzeroqueuedatacenternetwork AT shahdevavrat fastpassacentralizedzeroqueuedatacenternetwork AT fugalhans fastpassacentralizedzeroqueuedatacenternetwork |