DL380 slow performance

[shiljith]

dl380 performance for oracle feels so bad.in my current test server for oracle datawarehous(hp proliant ML110) only taking 3 hours to load full database but new server taking 5 Hours to complete even after i have removed raid controler.

what could be the reason for it.

new server is with 4GB of RAM and old server is 1.5 DB of ram.

[Gopi Sekar]

check the io wait time and io performance. check your dmesg and free command outputs and may be post it here for us to take a look.

if the bad performance continues even after removing raid controller then it is most likely the CPU and RAM related issue (mostly IO related).

Regards,
Gopi

[Steven E. Protter]

Oracle's storage guidelines:

Data/Index/redo : Raid 1.
Other stuff: Raid 5.

If you are loading your data on to RAID 5 storage, and the raid parity is 9 then every single write is being done nine times.

That has a significant(like a hammer) impact on db performance.

SEP

[shiljith]

hi SEP

i have removed the raid,now 3 HDs are just scsci drivers. its not only problem with performance,our datawarehouse fully truncating all the data and reloading all day.it is taking exactly double the time of normal servers.

shiljith

[Gopi Sekar]

I am not sure whether oracle uses buffers and shared memory/semaphore. But it would be better to check on that.

Regards,
Gopi

[Ross Minkov]

I'm just curious... How did you remove the RAID controller from the ProLiant DL380? It's an intergrated controller... not a PCI card. Are you using the internal bays to put your disks in or external storage shelf/enclosure?

-Ross

[Uwe Zessin]

I'm curious, too:
what is meant by "raid parity is 9" ?

[Steven E. Protter]

I'm the wrong guy to answer the parity question. I learned it from our disk array provider during our March 2005 disk meltdown disaster.

Correct term is Raid 5 parity 9 or 9 bit parity, not sure.

What it means is that every stripe goes across 9 disks on the SAN. To go totally off topic, we were told we lost our data the array failed because the parity exceeded the array's channels to write the data.

Surely someone will provide a better explanation. The default for Raid 5 is as I'm told parity 9, which results in a minimum of 9 writes to different disks for every write.

Back to the issue at hand, since RAID is not a factor, I'd check to make sure there are not a lot of scsi devices in the chain and that everything is terminated properly.

SEP

[Vitaly Karasik_1]

- What about just copying/writing big files?
- Do you have the same linux/kernel on both servers?

[Vitaly Karasik_1]

- and what is LINUX version on your DL380?

[Franco_24]

Hi,

Please consider this post:
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=952802

[Jean-Luc Oudart]

Hi,

I don't think the amount of RAM will matter for that type of upload (do you use sqlldr what mode).
If you use exact same method between ML110 and DL380, what are the other differences on the hardware ? nb disks , disk speeds, disk controlers ,...
could you post the information ?

Regards
Jean-Luc

[shiljith]

i had spend enough time to make it work,but it didn't. so i am already changed the server.

thanks to all

shiljith