On 09/04/2013 04:35 AM, Alfredo Rezinovsky wrote: > In the page reccomends reiserfs. So I tried to set up a squid for > 200Mbps squid using reiser and squid 3.3 > > I had a problem, every five to ten minutes, the system freezes and the > bandwith drops to a half for about 10 seconds. Yes, this is typical for either untuned or overloaded ext2 and ext4 file systems as well. > I tried tuning squid in many ways. And I noticed reiser flushes the > buffers all at a time, freezing the I/O, I suppose even the network I/O. Yes. AFAIK, _any_ Linux process might be frozen indiscriminately when fs is running behind its flushing schedule. > The solution was to reformart the cache disks using ext4 (noatime an > so). Now the proxy runs smoothly. In our tests, it was easier to configure ext2 to do the right thing with respect to flushing than ext4, but we were not using ufs-based cache_dirs and YMMV. Also, if you are using ufs-based cache_dirs then you are likely to start seeing similar overload/flushing effects if your Squid load increases. Besides fs tuning (to avoid excessive accumulation of unflushed data), the key is in limiting disk traffic, which ufs-based cache_dirs cannot do (yet?). On the other hand, ufs-based systems are not meant for performance-sensitive environments anyway :-). Cheers, Alex.
