Tuning RED for Web Traffic: RED Considered Harmful?
- A talk given by
at Sprint Advanced Technologies, Burlingame, CA,
in March 2000.
The IETF is promoting deployment of RED in Internet routers as an instance of
active queue management for congestion avoidance. While the effect of RED on
network-centric measures of performance has been well-studied, its effect on
application-level measures of performance has not.
This talk presents the results of a study of the effects of RED on the
performance of Web requests, specifically, response time for HTTP
request-response transactions. We have empirically evaluated RED across a
range of control parameter settings and offered loads in a laboratory setting.
Our results show that:
- Contrary to expectations, compared to a FIFO queue, RED has a minimal
effect on HTTP response times for offered loads up to 90% of link capacity,
- Response times at loads in the range 0-90% are not substantially effected
by RED control parameters,
- Between 90% and 100% load, RED can be carefully tuned to yield
performance somewhat superior to FIFO, however, response times are quite
sensitive to the actual RED parameter values selected. In particular
"recommended" RED settings produce poorer performance than FIFO and
optimal settings are non-obvious and were only arrived at through extensive
trial and error experimentation.
- Between 90% and 100% load, where RED has the potential to provide better
performance, performance becomes a subjective measure. For loads in this
range there exists a trade-off between improving response times of
short-lived connections and improving response times of long-lived
connections. Both cannot be optimized simultaneously.
- In such congested networks, RED parameters that provide the best link
utilization produce poorer response times.
We conclude that for links dominated by web traffic, RED appears to provide
no clear advantage over FIFO for end-user response times. Moreover, given
the lack of engineering practice to guide the setting of RED parameter values,
and an existence proof that "reasonable," RED parameters values can result in
poorer performance than FIFO queuing, it is possible that widespread RED
deployment may cause more harm than good.
slides for this talk.
Tutorials, short courses, conference presentations, and colloquiums page.