ONE LARGE I/O PROCESS SLOWS DOWN ENTIRE SYSTEM

[dblarry]

Hi,

We run most of our Oracle databases on Sun Solaris. We are now starting to run them on AIX (5L).

We notice that when a single write-intensive process runs on AIX (whether it be TAR, FTP, database backup , the entire box slows down. Is there something vastly different about I/O on AIX vs. SUN ? It seems like the I/O to one process becomes a priority and other processes don't get much priority.

Any general recommendations you can make? The boxes from AIX and Sun are of similar size (six cpus, 16 gig memory). Files system access is via SAN -- JFS (AIX) VERITAS (SUN).

Let me know if you have any recommendations.

Thanks

 

[Chapter11]

The general recommendation would be to make sure your I/O is effectively distributed across as many physical disks as possible. There are various other tricks that can be implemented as well, but it will boil down to specifics.

 

[bi]

I agree with Chapter11. I have had the exact opposite experience with I/O performance between AIX and Solaris. The problem with the Solaris boxes is that they are old, the disks are slow, there aren't many I/O cards. The AIX boxes are newer, the disks are newer and more numerous, there are multiple I/O cards to spread I/O across as well.

What kinds of disks are you using on the two different systems?

 

[dblarry]

Actually, we are using the same type of disks on the SAN with EMC for both the AIX box and the Sun box. We feel that the disks or striping of disks is not the problem. Seems like the AIX box just doesn't have the throughput to handle many concurrent users. Anyone think JFS vs. Veritas could be an issue ?

One other note: If am doing an Oracle import and I am the only one on box - this AIX box does fly -- importing a 200 gig database much faster than the SUN. However, when multiple people are on the box -- they would experience tremendous slowness if I do a large write process. On SUN the import is slower -- but has little impact on the other users.

 

[MoshiachNow]

dblarry,

Please post the results of the following taken during the high write load on the AIX:

1. vmstat 2 20
2. topas (one screen shot)
3. lsdv -Csscsi
4. lsattr -El hdiskx (of the disk to which the write occurs)

Thanks


"Long live king Moshiach !"

http://www.7for70.com/

 

[sbix]

What does it say a "vmstat 3 20" while this program runs?
How big are the values in the "wait" column?