poor performance on big machine
dillon at apollo.backplane.com
Mon Jul 30 07:54:25 PDT 2007
:I just continued building and noticed that we seem to have a horribe hdd performance:
:Disks ad4 ad6 acd0 cd0 pass0 md0
:KB/t 0.00 44.27 0.00 0.00 0.00 0.00
:tps 0 186 0 0 0 0
:MB/s 0.00 8.03 0.00 0.00 0.00 0.00
:% busy 0 100 0 0 0 0
:A lot of processes are in states getblk, wdrain, biowr, biord. Not stuck, but real slow. Everybody is pounding on ad6.
This usually means the disk is seeking around a lot. There are several
reasons this might be happening. The primary culprit is usually that
the machine is being forced to page or swap due to insufficient memory
for the process load. Certain types of less cachable loads can also
result in a great deal of disk seeking.
Hard drives are only 'fast' when doing linear reads and writes. A
seek on a hard drive usually takes 3-8 mS depending on the drive. Most
seeks are hidden or absorbed by either the hard drive's on-board track
cache or by the operating system's buffer cache but certain types of
loads can defeat both.
:For some reason "test" was also several times in "biowr". First, I don't understand why test is used and not the builtin, but whatever. What's the business of test reading something? Might that be urandom and stackprotector? When else would test read a block?
Is the machine swapping? Look at the SWAP PAGER stats with systat -vm 1.
If it is swapping heavily for long periods of time this is usually an
indication of not having enough memory for the process load. DragonFly
is pretty good at keeping things running even during periods of heavy
paging loads but something still has to give and it usually winds up
being the disk efficiency.
<dillon at backplane.com>
More information about the Kernel