Why flow-completion time is the right metric for congestion control
By Nandita Dukkipati, Nick McKeown. CCRV, 2006
358 citations recorded by the source corpus.
Subfields: Computer Networks and Communications
View Paper: https://www.semanticscholar.org/paper/a3252cb71056d4fc0a6da6cf0c607eb46e06ed7e
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Networks and Communications
- Anirudh Sivaraman
- Brighten Godfrey
- Mohammad Alizadeh
- Hari Balakrishnan
- Michael M. Mitzenmacher
- Keith Winstein
- George Varghese
- Vimalkumar Jeyakumar
Citations in the corpus
- Adaptive forwarding in named data networking (2012, methodology): “suggest to use flow completion time as the metric for congestion control and propose RCP [26] to help short flows gain their fair share of the bandwidth more quickly.”
- CONGA: distributed congestion-aware load balancing for datacenters (2015, methodology): “Similar to prior work [20, 5, 13], we use the flow completion time (FCT) as our main performance metric.”
- PINT: Probabilistic In-band Network Telemetry (2020, methodology): “Application Description Measurement Primitives Per-packet aggregation Congestion Control [22, 29, 40, 46] Congestion Control with in-network support timestamp, port utilization, queue occupancy Congestion Analysis [17, 38, 57] Diagnosis of short-lived congestion events queue occupancy Network Tomogr”
- An experimental study of the learnability of congestion control (2014, background): “Many such metrics have been proposed, including alpha-fair throughput [26], flow completion time [8], throughput-over-delay [22], or measures based on a subjective opinion score [13].”
- Swift: Delay is Simple and Effective for Congestion Control in the Datacenter (2020, neutral): “Explicit Network Feedback including INT: XCP [27], RCP [17], and D3 [54] rely on switches to feedback rates to end-hosts.”
- Finishing flows quickly with preemptive scheduling (2012, methodology): “TCP, RCP [10], ICTCP [22], and DCTCP [3] approximate fair sharing, dividing link bandwidth equally among flows.”
- Reproducible network experiments using container-based emulation (2012, background): “FCT [45] The relationship between Flow Completion Time (FCT) and flow size is not ideal for TCP; small flows take disproportionately long.”