Analyzing the performance of TCP flows under LAS schedulers

The networking community has focused mainly on Processor Sharing (PS) policies in routers in
order to establish QoS guarantees. As of today, PS scheduling has not been deployed in routers and
the Internet still uses FIFO scheduling that offers only a best-effort service. Also, Internet traffic
exhibits a high variability in terms of the flow sizes: Most of the flows are short, while more than
half of the bytes are carried by a few largest flows. Given these observations, we recommend Least
Attained Service (LAS) scheduling in routers, which favors short flows by giving the highest priority
to the first packets of each flow. As a result, LAS improves the overall user perceived performance.
Using simulation we show that under LAS scheduling, as compared to FIFO, the transmission time
and loss rate for short TCP flows are significantly reduced, with a negligible increase in transmission
time for the largest flows. We also show that the improvement seen by short TCP flows under LAS
is mainly due to the way LAS interacts with TCP algorithm in the slow start (SS) phase, which
result in shorter round-trip times and very low packet loss rates for short flows. We also investigate
the performance of long-lived TCP and UDP flows under LAS and compare the results to FIFO
scheduling.