Using synchronized clocks to improve temporal fairness, congestion control, and load balancing in datacenter networks
Abstract/Contents
- Abstract
- Datacenter networks have long operated under a clockless assumption: network devices cannot rely on timestamps from local clocks for global network-wide decisions. This assumption has served network architects and distributed systems developers well for the past 30 years. However, with the emergence of cheap, scalable, and host-based clock synchronization solutions with nanosecond-level synchronization accuracy, it is now possible to relax this assumption. In this dissertation, I will explore the new potential that accurate clock synchronization unlocks in datacenter networks. First, I will present CloudEx, a fair-access financial exchange system in the public cloud. CloudEx provides fairness for order processing and market data dissemination by compensating for the variable delays in the cloud using accurately synchronized timestamps. CloudEx has been used as a research tool for trading infrastructure and algorithms and has also been deployed as a teaching tool in a financial technology course at Stanford. Second, I will discuss On-Ramp. On-Ramp uses synchronized clocks to measure accurate one-way delays as a signal for congestion, then enforces precise pausing of packet transmission on flows when it senses congestion. I will present my contributions on testing On-Ramp in Meta to reduce the tail latency of a production service by 2.5-3x for the same throughput. Last, I will present Reactive Subflow Spraying (RSS), a multipath and reactive load balancing algorithm that uses synchronized clocks to measure accurate one-way delays as a signal for network load imbalance. I show how the performance of RSS in terms of flow completion times is within 5-10% of the optimal performance and is 2-3.5x better than other mechanisms in asymmetric setups, e.g., in the presence of link failures.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Ghalayini, Ahmad |
|---|---|
| Degree supervisor | Prabhakar, Balaji, 1967- |
| Thesis advisor | Prabhakar, Balaji, 1967- |
| Thesis advisor | Kabbani, Abdul |
| Thesis advisor | Rosenblum, Mendel |
| Degree committee member | Kabbani, Abdul |
| Degree committee member | Rosenblum, Mendel |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Department of Electrical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Ahmad Ghalayini. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/jb363hk4434 |
Access conditions
- Copyright
- © 2023 by Ahmad Ghalayini
- License
- This work is licensed under a Creative Commons Attribution 3.0 Unported license (CC BY).
Also listed in
Loading usage metrics...