I have configured two servers each with 32 GByte RAM, 2x Xeon 2.4 GHz, 4x 480 GByte Intel SSD on a LSI MegaRAID SAS 9261-8i controller. They should be identical, however one of them has a very poor performance for operations like 'du' and 'find'.
For example, a 'du' command on ten internal subversion directories with 1000 files each takes 31 seconds on the slow system, but returns instantly on the other one. Files are of course identical.
root@server26:/var/tmp# date; du -skh /home/web0000/svn/db/revs/10?; date Wed Jan 16 22:01:06 CET 2013 886M /home/web0000/svn/db/revs/100 2.0G /home/web0000/svn/db/revs/101 766M /home/web0000/svn/db/revs/102 317M /home/web0000/svn/db/revs/103 533M /home/web0000/svn/db/revs/104 937M /home/web0000/svn/db/revs/105 486M /home/web0000/svn/db/revs/106 961M /home/web0000/svn/db/revs/107 1.3G /home/web0000/svn/db/revs/108 1.5G /home/web0000/svn/db/revs/109 Wed Jan 16 22:01:37 CET 2013
I have also tried the same command on a different server with conventional hard disks (RAID1), the command completes in less than one second. That server has been rebooted to make sure nothing is in the file system cache.
Similar problems when running 'du' or 'find' on /usr or /var.
The read/write performance for a single large file seems to be ok:
root@server26:/var/tmp# dd if=/dev/zero of=largefile bs=102400000 count=100 100+0 records in 100+0 records out 10240000000 bytes (10 GB) copied, 39.4998 s, 259 MB/s
Read speed is about 500 MB/s. But I have not the faintest idea why a simple 'du' or 'find' command is so slow. Please advise.