On 02.08.2009, Adrian Chadd wrote: > > 2. Format /dev/sdb1 with "mkfs.xfs -f -l lazy-count=1,version=2 -i attr=2 -d agcount=4" > > 3. Mount it afterwards using "rw,noatime,logbsize=256k,logbufs=2,nobarrier" in fstab. > > 4. Use cfq as the standard scheduler with the linux kernel > Just out of curiousity, why these settings? Do you have any research > which shows this? I'll see what survived of all the testings I did. Had a harddisk crash in the meantime and I'm not shure if I have saved the results. These are the parameters related to XFS which gave the best latency and highest throughput on my system with an OCZ SSD (cacheless). I "benchmarked" also ext2/3/4 and jfs, and did it separately for cfq, dl, anticipatory and the noop elevator. I had also posted some of these results on the lkml, here's one link I found by a quick search, which shows how cfq outperforms dl nearly by a factor of 10: http://lkml.org/lkml/2009/4/8/664 I did a while : ; do time sh -c "dd if=/dev/zero of=bigfile bs=8M count=256 ; sync; rm bigfile"; done to put the system under heavy I/O, and measured the responses with the fsync-tester. The results were noticeable, also without measuring. I did also a lot of bonnie++ runs: bonnie++ -u htd:users -d /mnt/test -s 16016m -m liesel -n 16:100000:16:64 > > (Btw: on my systems, squid-2.7 is noticeably _a lot_ slower than squid-3, > > if the object is not in cache...) > This is an interesting statement. I can't think of any specific reason > why there should be any particular reason squid-2.7 performs worse > than Squid-3 in this instance. Forget about that. Yes, that's "black magic", and it's NOT a general statement, but an observation. Was just interested in how squid-2 performs on an old P4 realted to squid-3. It's quite possible that I did something wrong which was the cause.
