On Tue, 11 Sep 2001, Eiji Kawai wrote:
> I have noticed another point about no surrogate test in webaxe-4.
> The how-to document at http://www.measurement-factory.com/docs/WebAxe-4/
> says that the dummynet on all the polygraph clients should be
> configured with the delay of 40ms and the packet loss rate of 0.05% on
> each way from/to these clients.
>
> However, the webaxe-4 document at
> http://www.web-polygraph.org/docs/workloads/webaxe-4/
> says that the delay and the packet loss rate should be 100ms and 0.1%,
> respectively.
The WebAxe-4 description on the Web is now corrected to say:
client side: 40msec delay and 0.05% packet loss in both directions
server side: no delays or packet loss
The reverse is true for PolyMix-4:
client side: no delays or packet loss
server side: 40msec delay and 0.05% packet loss in both directions
> I did the no surrogate tests with both these environments and
> found that the response time experienced by polyclt is about 500ms
> with 40ms dummynet delay, however, it is about 700ms with 100ms
> delay.
>
> The how-to document describes the response time should be about
> 0.53 seconds, and I cannot determine my result is incorrect by
> some miss-configuration or not. Do you have some suggestion about
> this result?
The 500-530msec response time with 40msec delays sounds about right.
Please note that server think time for WebAxe is ~300msec. The
additional 200-230 msec are the effect of per-packet delays (80msec
for each transaction plus ~160msec for each new TCP connection),
packet loss, and client/server overheads... Please let me know if you
disagree.
To summarize, use 40msec packet delays, 0.05% packet loss, and expect
500-530msec response times in a no-surrogate tests.
Thanks,
Alex.
This archive was generated by hypermail 2b29 : Mon Feb 06 2006 - 12:00:19 MST