Abstract
To isolate service queues in data centers, recent solutions assume Explicit Congestion Notification (ECN)-enabled end-hosts. However, this assumption is not valid in multi-tenant environments where tenants own the network stack. This letter presents BarberQ, a multi-queue management scheme that isolates service queues without dependency on transport protocols. At the heart of BarberQ lies dynamic buffer sharing through packet eviction. Our benchmark results demonstrate that BarberQ not only achieves minimum guarantees and work conservation, but also provides low latency.
| Original language | English |
|---|---|
| Article number | 8826238 |
| Pages (from-to) | 1949-1952 |
| Number of pages | 4 |
| Journal | IEEE Communications Letters |
| Volume | 23 |
| Issue number | 11 |
| DOIs | |
| Publication status | Published - 2019 Nov |
Bibliographical note
Funding Information:Manuscript received June 20, 2019; revised August 9, 2019; accepted September 2, 2019. Date of publication September 6, 2019; date of current version November 11, 2019. This research was partly supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science and ICT) (No. 2019R1A2C2088812), Next-Generation Information Computing Development Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT (No. 2017M3C4A7083676). The associate editor coordinating the review of this letter and approving it for publication was W. Cerroni. (Corresponding author: Wonjun Lee.) The authors are with the Network and Security Research Laboratory, School of Cybersecurity, Korea University, Seoul 02841, South Korea (e-mail: [email protected]; [email protected]). Digital Object Identifier 10.1109/LCOMM.2019.2939814 Fig. 1. Goodput and queue length when two services share the bottleneck. Service 1 generates a flow at the beginning and service 2 generates 8 flows at 0.5 seconds. Although each service has the same weight and the bandwidth is expected to be guaranteed by packet schedulers, service 1 cannot grab a half of bandwidth due to excessive buffer occupancy of service 2.
Publisher Copyright:
© 1997-2012 IEEE.
Keywords
- Data center networks
- clouds
- multi-tenancy
ASJC Scopus subject areas
- Modelling and Simulation
- Computer Science Applications
- Electrical and Electronic Engineering