Alex Rousskov writes:
> Since even a half empty the buffer is flushed automatically within a short
> delay (as far as I understand), it is hard to notice the influence of the
> buffer for misses on _slow_ outbound connections.
Do you know how Squid manages to flush the output buffer automatically?
As long as it's on the standard TCP stack, I think it's very hard...
Is it something todo with "non-blocking IO"? I'm not very familiar
with network programming...
> http://www.cs.ndsu.nodak.edu/~rousskov/research/cache/squid/profiling/stats/gindex.all.wd/gbag.distr.cortvsz.html
>
> Look for a "jump" in proxy response time at 16, 32, or 64KB point (those
> are typical values for the socket buffer size, maximum is 64K). Visible on
> most of the proxies we have studied. The collection also has similar
> graphs for _total_ response time.
This is a very interesting result. Do you have any published paper on
your results? If you have let me know. Thank you very much.
-- H.TaniguchiReceived on Thu Dec 04 1997 - 09:19:02 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:50 MST