Hi all,
Hope someone might have an idea what's happening.

If I copy the same 680meg file within the same partition to another folder repeatedly (testing load average and copy times) about the third time the job finishes in about 10 seconds and drives the CPU load average up to around 7 as well as hang the box for a few seconds (around 10 sec.).

I was able to reproduce this on a seperate but hardware/software identical server.

RH7.2 (2.4.7-10)
Compaq DL580
4 X 700mhz Xeon
2 gig RAM
cciss 5300 array controller (RAID5)

Any thoughts welcome.

Ed

i dont know how/if raid fits into this but it sounds like your hard disks have not got dma enabled, what sort of output do you get from

hdparm -Tt /path/to/harddisk/device/file

If you are using a journaled file system (ext3, or reiserf's) you'll need time for the journaling to take place. Moreover, the internal buffers become full and need to be flushed. When you do a steady copy, you don't give the system time to do its thing in the background, so it moves the operations to a higher priority to get done for safety's sake.

Cheers--
Charles

Not being familiar with hdparm I looked up the man for it.

a command line interface to various hard
disk ioctls supported by the stock Linux ATA/IDE device
driver subsystem.

I figured I would try it anyway and got this:

]# hdparm -Tt /data/testDBTest
/data/testDBTest not supported by hdparm

I figure this is because the 5300 is a scsi controller and wouldn't be compatible with hdparm.

Any other tools like that one that might work with SCSI?

AH! the journaling filesystem overhead... forgot about that. Using EXT3 and probably pushing it. This would make sense.

Thanks for the help.

Ed