Hi!
On Apr 6, 2005 9:57 PM, Carlos E. R.
The Monday 2005-04-04 at 11:35 +0300, Hugo wrote:
Any ideas? Any pointers to check? We really do not have any idea what is slowing down the server?
Just one idea. You can check, locally, the transfer speed of the raid setup. Perhaps you can also setup a non-raid partition, export it through nfs, and check the speed from the client. Then you could know what is the problem: one, two, or the combination.
The problem is local (did some runs with bonnie++). And we've been able to correct the situation a little. We managed to get read speeds to the same decade than what 3ware gives out. Write speeds are still a little too slow, but much much better. No major source of problem was identified. (I actually do not know the details anymore... this was out of my leaque :-) but read-ahead settings needed to be fixed in all the right places. But still the loads seem to be up. The situation now seems to be so that if there are lot's of processes, then the processors do not spend time in i/O waits, but if there are not that many processes and processors can idle, then they will still spen uptp 80% in I/O wait (not idle). The setup we have is - 8 disks in SATA-RAID visible to the SuSE as one large Logical Disk. - /boot is on it's own partition on the RAID - everything else is configured as one LVM - on that LVM we have separate partitions for / (root), /swap, /var and /loca (almost all of the rest) Is there something fundamentally wrong with this? Or should this kinf of setup work OK? -- Hugo.