Recent studies have observed that large data center networks often have a few hotspots while most of the network is underutilized. Consequently, numerous data center network designs have explored the approach of identifying these communication hotspots in real-time and eliminating them by leveraging flexible optical or wireless connections to dynamically alter the network topology. These proposals are based on the premise that statically wired network topologies, which lack the opportunity for such online optimization, are fundamentally inefficient, and must be built at uniform full capacity to handle unpredictably skewed traffic. We show this assumption to be false. Our results establish that state-of-the-art static networks can also achieve the performance benefits claimed by dynamic, reconfigurable designs of the same cost: for the skewed traffic workloads used to make the case for dynamic networks, the evaluated static networks can achieve performance matching full-bandwidth fat-trees at two-thirds of the cost. Surprisingly, this can be accomplished even without relying on any form of online optimization, including the optimization of routing configuration in response to the traffic demands. Our results substantially lower the barriers for improving upon today's data centers by showing that a static, cabling-friendly topology built using commodity equipment yields superior performance when combined with well-understood routing methods.
|Original language||American English|
|Title of host publication||SIGCOMM 2017 - Proceedings of the 2017 Conference of the ACM Special Interest Group on Data Communication|
|Publisher||Association for Computing Machinery, Inc|
|Number of pages||14|
|State||Published - 7 Aug 2017|
|Event||2017 Conference of the ACM Special Interest Group on Data Communication, SIGCOMM 2017 - Los Angeles, United States|
Duration: 21 Aug 2017 → 25 Aug 2017
|Name||SIGCOMM 2017 - Proceedings of the 2017 Conference of the ACM Special Interest Group on Data Communication|
|Conference||2017 Conference of the ACM Special Interest Group on Data Communication, SIGCOMM 2017|
|Period||21/08/17 → 25/08/17|
Bibliographical noteFunding Information:
We would like to thank Ratul Mahajan for his insights on the limitations of the restricted topology adaptation model. We are also grateful to our colleagues who provided helpful feedback on this work, including Monia Ghobadi, Torsten Hoefler, George Porter, Marcel Schneider, Laurent Vanbever, and the anonymous SIGCOMM reviewers and shepherd. Asaf Valadarsky is supported by a Microsoft Research Ph.D. Scholarship. Michael Schapira is supported by the PetaCloud Consortium.
© 2017 ACM.
- Data center