RS/6000 B50 server slow (wait 97%)???

[elmotec]

One of our B50 servers (AIX 4.3.3) is pretty slow. I think this is because of the system is waiting (90 - 99%) for nothing. How can help? Output of "topas" is:

[tt]
Kernel 1.7 |# |
User 1.2 | |
Wait 97.0 |########################### |
Idle 0.0 | |


EVENTS/QUEUES FILE/TTY
Cswitch 101 Readch 12075
Syscall 533 Writech 8837
Reads 34 Rawin 33
Writes 15 Ttyout 683
Forks 0 Igets 0
Execs 0 Namei 18
Runqueue 0.0 Dirblk 0
Waitqueue 0.0

PAGING MEMORY
Faults 33 Real,MB 511
Steals 0 % Comp 30.0
PgspIn 0 % Noncomp 70.0
PgspOut 0 % Client 0.0
PageIn 0
PageOut 0 PAGING SPACE
Sios 0 Size,MB 1024
% Used 15.7
% Free 84.2

Interf KBPS I-Pack O-Pack KB-In KB-Out
en0 3.9 14.2 9.8 1.2 2.7
lo0 0.4 0.8 0.8 0.2 0.2

[/tt]

 

[courcol]

What about disk/IO activity ?

 

[elmotec]

Disk I/O differs a lot. The next data are just two snapshots:

[tt]
Disk Busy% KBPS TPS KB-Read KB-Writ
hdisk1 50.6 388.7 70.6 40.0 348.7
hdisk0 22.5 141.0 27.5 0.0 141.0

Disk Busy% KBPS TPS KB-Read KB-Writ
hdisk1 6.0 28.0 6.0 14.0 14.0
hdisk0 0.0 0.0 0.0 0.0 0.0
[/tt]

hdisk0 is the 9Gb system disk with AIX 4.3.3
hdisk1 is the a raid-5 with 4x36Gb for data.

Thanks.

 

[courcol]

Seems to be few activity on hdisk1 (slow old disks ?), and RAID5 isn't the fastest of RAID level, mainly when it's soft RAID with poor disks or uneffective controllers...

If there is only 1 process running on your server (like 1 database process waiting for IO access), with not enough cache, the only thing your CPU has to do is waiting for this IO (I've got quite the same problem on a StorageTek disk array, with Oracle database).

 

[elmotec]

hdisk1 is an SSL based hardware RAID5 configuration. We use the same in an other server and that one is blazing fast!

The second point can be the real problem. On the RAID5 is an Oracle database stored. How can you increase the cache you mentioned? Or do you mean the processor cache itself?

 

[aixat]

Hello,

search on this forum on oracle and avadh. you will find hit that has tuning for oracle.

 

[courcol]

Concerning Oracle, you can play with db_block_buffer (and check that db_block_size match your filesystem block size at the same time (lsfs -q)) in the init<sid>.ora file.

You can also try to defragment the filesystem, and verify the placement of blocks on disk (lsvg -p <volume group>), center is better.

About increasing Aix filesystem cache, as I rather play with Solaris boxes, I'm still looking for how to tune it. Tell you more as soon as possible

 

[courcol]

Great, check article &quot;Ambient temperature on RS6000 and SSA enclosures&quot; (sounds like nothing to do with disk IO access ?), there is good stuff on tuning aioserver performances.

Thanks, alxat