hi there,

maybe someone has a suggestion about this..

.. i have a pretty fast server (dual xeon 3.0 ghz, 4 gb ram, lindenhurst chipset) and equipped it with a 3 ware 9500-s4 raid controller. it uses 4 sata-drives of 120 gb capacity in 2 arrays (raid 1). array one is system, webserver, database. array 2 is currently used to keep a longer history of database dumps. but shall be for the database only, but later.

the system on the server is debian sarge with kernel 2.6.10.

so far it is running fine. but when it comes to jobs with a high amount of disk i/o i'm not sure if everything is right.

for example, when i do a dump of an 1.8 gb database and gzip it. that actually does not freighten the cpu's much. the job runs via nice with a moderate load. but the wa-load is very high and while the job runs the server is nearly unaccessible. this goes on for ~ 6-10 minutes.

it gets even worse when i try to import a dump of that size into the database. the system virtually freezes and sits there with 100% wa-load. the cpu's are near zero all the time and this goes on for 3 hours.
i import the same dump in ~20 minutes on a lousy xp2000 server with a cheap onboard raid1 under windows 2003 with no database optimizations.

now that tells me something is really wrong with the xeon server. it behaves like when you have some small and slow ide harddisks under windows and forgot to turn on the DMA. so all the PIO action occupies the system.

any suggestions on this?

i looked into the logs.. nothing special as far as i can tell. the raid card did not give any errors (checked with CLI).

Install the 3dm web server. You can increase I/O and also find out if anything is amiss with your drives aka you have a dead drive. You can also do a verify RAID and scan for media errors(bad clusters).

You will not be able to detect media errors on boot. You need to use the utility. The bios can only detect complete drive failures.

Honestly, it sounds like one of your drives in the array is failing. And they aren't Raptor 10K RPM speed demons. Just 7200 IDE drives. So any deversions such as one slow failing drive would screw it up noticeably.

i got the CLI util on the server. all the output says nothing about errors. log is empty, all drives and units are OK.

will perform a hard disk check. but i guess this will block the server very badly. have to pick the right time.

however harddisk write cache is off at the moment. this is because the server is still tested for stability.

you think this all could be poor harddisk performance? hmmm.. that could be for the usually high wa load. i don't expect wonders from the drives here.

otoh for the mysql operations that last for hours and hours with next to no cpu load the cause must be something else. i hope the harddisk check brings something to light.

thanks for your reply so far.

You must manually initialize a media/verify scan. It will not find errors unless it is looking for them. I have not done this via the CLI yet. But, I have done this via the http interface.

My suggestion- how I setup my database servers is this:

75 GB Raptor 10,000 RPM SATA X6
2 Drives @ RAID 1 /boot and userland
4 Drives @ RAID 1+0(10) MySQL

You can get a 3ware 4 port and 2 port SATA card, or get one 8 port card if you have a 64bit PCI bus.

In your situation, i really think one of your drives are having errors aka bad sectors- and your hard drives a freezing/catching when they try to read/write to it.