Slow Performance on Unix/Oracle

Hello

We seem to be getting slow runtimes (20 hours) time when the followong occurs:

Scenario:
We run runbatches on our FAST Unix server that access a lot data (i.e. select * from proddta.F0911 ) from our Oracle 9.2.0.3 database. This is true even if nothing else is running on our Unix application server.
Even more interesting is if we run this same job locally on a Fat client on a SLOW windows PC, the job runs rather fast (2 hours).

TESTING:
We have tested running "Select * from proddta.f0911" using Sqlplus from both the Unix server and the Pc that has the fat client and when this occurs they both run equally fast (about 2 hours). This leads to think it is not in our netwok or sqlnet connection from the Application Server to the Database Server.
We have done oracle traces for the this job running on both scenarios (PC and Unix Server). These traces indicate that Oracle is waiting much longer for a message from the client process for more data (SQL*Net message from client) on the Unix Server than oracle waits for the same job running on the PC. The total time difference between the PC and Unix Server for "SQL*Net message from client" messages is about the total difference in runtimes (17.4 hours).

Possible Reasons:
This makes me think that the unix server is:
A) slow when it comes to processing JDE runbatches do to the JDE code as written.
B) We have some configuration issues

In our JDE.INI on our UNIX (See Below) server we have the folowing parameters set up. Could this be slowing our processing of runbatches on our Unix App server that access large amounts of data?

[DB CACHE INFORMATION]
ODBC Tables=50
Maximum Request Cache=50
Library Cache=15


If anyone has any ideas about how to address this, We woud be most appreciative.

Thank you.
Googers

OneWorld XE SP22_E1/Unix Solaris 8/oracle 9.2.0.3

Googers

I looked at both my client and enterprise jde.ini file and I don't have the DB Cache paragraph at all. I don't know if it's necessary in yours. You have a different configuration than me.

Patty

We have similar issues with Oracle and Unix at my site.

Here are some tweaks to the JDE.INI that seemed to have helped us at our site. I suggest you try them in your test enviroment first.


Under [DB SYSTEMS SETTINGS] add
OracleServerHandleReuse=0

Here's our [DB CACHE INFORMATION]
ODBC Tables=50
Maximum Request Cache=50
Library Cache=15
DataCaching=1

Comment out ";checkKrnHealth=1000" under the [JDENET] paragraph

Here's our [MEMORY DEBUG]
;Frequency=10000
;Full=1

set "ipcTrace=0" under [JDEIPC]

Under [JDENET] set "netTrace=0"
and comment out ";checkKrnHealth=1000"

In addition we found that we had to allocate more Unix resources to JDE, we added more physical memory, tweaked /etc/system and the corresponding values in JDE.INI to take advantage of the memory.

Thanks,
BennyC

XE/SP20_G1, Websphere 3.5, JAS and Fat clients, Solaris 8, Oracle

Thank you for the help

In addition to Benny's suggestions I would add the following:

1. JDE.INI
Under [LOCK MANAGER]
Change RequestedService from TS to NONE, per oti-99-0024.

2. Oracle
Is OneWorld using IPC to communicate to Oracle or TCP? Should be IPC on the server.

3. We have always observed that most PCs can run UBEs faster than our HP DS-390 box. Given that the E.S. has a lot more tasks to perform simultaneously than a PC its not so surprising. Think of an Enterprise Server as a powerful Budweiser Clydesdale rather than an Arabian race horse. The Clydesdale may not run as fast but it can haul a heck of a lot more beer than the race horse. Not to say that a 10:1 ratio isn't surprising, I've never seen that here.

Good luck and keep us informed on your findings.