James R Grinter wrote:
> oh, this is a shame - I've obviously not been following closly enough,
> recently. Is anyone looking at implementing in a different manner (eg
> multiple processes to schedule the reads/writes in a way akin to the
> dnsservers)?
Don't worry to much.
The current async-io code is there, but the design of it can't be
brought much further than it is. The current code quickly makes the CPU
a bottleneck.
Yes, there are activities in finding a suitable replacement, both for
the I/O and the actual storage.
What Alex refers to is that I have now officially ended my attempts in
getting the current async-io code to perform better and more stable in
high workloads, and will instead spend my time on a different approach
to the problem, when time permits that is.
The net effect is that the async-io in Squid 2.3 may not perform as well
or stable a in patched Squid-2.2, but in time there will be other
alternatives which hopefully perform better, even thou it may take some
time before everything is in place.
-- Henrik Nordstrom Squid hackerReceived on Sat Oct 02 1999 - 15:49:36 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:48:42 MST