topic Re: CPU or Network Bottleneck in Operating System - HP-UX

CPU or Network Bottleneck

LudekC — Fri, 02 Mar 2007 05:19:42 GMT

Hello,
Good day,
configuration:
HP-UX B.11.11 not patched
Oracle 9.2.0.4
application server rp5470 (4 CPU)
database server rp2470 (2 CPU)

On application server GlancePlus (gpm) display Alarms:
CPU Bottleneck probability = 100
Network Bottleneck probability = 70

CPU LOAD avg sometimes more then 4 or 5 or 6
(1 or 2 day in moon when many people work)

I attach some system snap.

Thank You Ludek

Re: CPU or Network Bottleneck

Karsten Löperick — Fri, 02 Mar 2007 08:34:17 GMT

Hello Ludek,

just my ideas.
It looks a bit like a hard-disk problem.
I assume agroa is the database server. Please check out your DB configuration ( on which hardisk/array are the data stored ). The disk c4t0d0 and c4t1d0 reporting often 100% busy. What is at this disks ? Are this disks or the array with this disks failure free. 100% busy time means also a longer avwait time ( time to wait before transfer process can work ) that can cause a higher CPU utilisation.
Have you allready checked the network statistic via lanadmin ? With lanadmin you can see also some lan error statistics.

Greetings

Karsten

Re: CPU or Network Bottleneck

Oviwan — Fri, 02 Mar 2007 09:03:40 GMT

Hey

I think also its the disk. make a snapshot of the DB and have a look at the Buffer hits. If they are <80% I recommend you to give more cache to your db.

Regards

Re: CPU or Network Bottleneck

LudekC — Mon, 05 Mar 2007 08:46:46 GMT

Hello,
I apologize, agroa server is application server. Name database server is agrod.
Disk c4t0d0 and c4t1d0 are disk where are application files, C-ISAM *.dat files - database used our cobol information system. Previosly all files are in *.dat, but now majority files are transformed into Oracle database. On disk are not errors. Between application and database server is 1GB net.
We have alarm CPU Bottleneck so I 1 moon ago change on net cards packet size MTU from 1500 (on Oracle defaul setings) on 9000 bytes (Oracle settings SDU=TDU=MTU=8192). I want test SDU = 9000 or 18000 nad TDU = 32000(default).
I attach snapshot from Oracle server.
Here is for example SQL:
Module: pwahl2.RTO@agroa (TNS V1-V3)
SELECT /*+ INDEX( ts ts_key) */ * FROM
agroagro1."TS"
Export file table TS have 166MB and execution 72,302 in 13 min?

Thank You Ludek

Re: CPU or Network Bottleneck

LudekC — Tue, 06 Mar 2007 03:02:03 GMT

Hello,
Oracle utility staspack not display WHERE clausule in SELECT /*+ INDEX( ts ts_key) */ * FROM
agroagro1."TS"
Is returned only 1 row.
Ludek

Re: CPU or Network Bottleneck

Karsten Löperick — Tue, 06 Mar 2007 07:01:26 GMT

Hello again,

meanwhile I dont think it is a Oracle problem. Please have a look to the lanadmin tool - specially check out the error rates. Are there any differences compared with MTU=1500. High CPU % and Harddisk % are also possible if NIC and HD Controller on the same PCI-Bus ( some Interrupt problems during the highspeed transmission ). Check out how is your system hardware configured. Please check also your router/switch - is there also enabled to serve jumbo-frames ?

Greetings

Karsten