Thanks for the tip Amos. I tried compiling my own version with BUFSIZ set to 32KB, but it didn't seem to help. The TCP buffer size on my system is 212992 bytes, I tried 64KB too, but that also didn't improve my situation. Aside from adjsting the read_ahead_gap, is there anything else I'm missing? I'm sort of out of my league here so I may just quit and wait for v4. ;) Thanks, Jamie >Sadly, that is kind of expected at present for any single client >connection. We have some evidence that Squid is artificially lowering >packet sizes in a few annoying ways. Used to make sense on slower >networks, but not nowdays. > >Nathan Hoad has been putting a lot of work into this recently to figure >out what can be done and has a performance fix in Squid-4. That is not >going to make it into 3.5 because it relies on some major restructuring >done only in Squid-4 code. > > >But, if you are okay with playing around in the code his initial patch >submission shows the key value to change: ><http://lists.squid-cache.org/pipermail/squid-dev/2016-March/005518.html> >which should be the same in Squid-3. The 64KB bump in that patch leads >to some pain so dont just apply that. In the end we went with 16KB to >avoid huge per-connection memory requirements. It should really be tuned >to about 1/2 or 1/4 the TCP buffer size on your system. >After bumping up that read_ahead_gap directive also needs to be bumped >up to a minimum of whatever value you choose there. > >HTH >Amos _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users
