OAS 10.1.3.1.0 and Scaling

[Terry Duffy]

Hi,

I am trying to find out if anyone is currently using 8.97 with OAS 10.1.3.1.0 on a windows 2003 Enterprise Edition OS and more than 80 users either spread over multiple jvm's or on a single jvm.

I am particular;y interested in whether you have amended any http parameters or OAS parameters to aid in performance or scaling.

We are having problems getting past 80 users, when the system starts to behave erratically. Oracle Support are slowly going through everything and from the jde erspective it all looks OK.

Any help / pointers greatfully received.

E811 SP1 / OAS 10.1.3.1.0 / 8.97.03 / Win2003 Enterprise Edition 8Gb Ram Web Server / Oracle 10g RAC Clustered database on Linux / Win2003 Enterprise Edition 8Gb Ram Logic & Batch Servers

 

[Sebastian Sajaroff]

Hi,

May you describe us your servers hardware specifications?

 

[Terry Duffy]

I thought that I mentioned it in the post, but I will expand on the specs below.

Web Server Config :-

Microsoft Windows Server 2003 R2 Enterprise Edition SP1
Intel(R) Xeon(TM) CPU 3.2Ghz (4 CPU's)
8Gb RAM PAE

One HDD 68.2Gb

Logic Server Config :-

Microsoft Windows Server 2003 R2 Enterprise Edition SP1
Intel(R) Xeon(TM) CPU 3.2Ghz (4 CPU's)
4Gb RAM PAE

Two HDD 68.2Gb & 48.9Gb

Batch Server Config :-

Microsoft Windows Server 2003 R2 Enterprise Edition SP1
Intel(R) Xeon(TM) CPU 3.2Ghz (4 CPU's)
2Gb RAM

Two HDD 68.2Gb & 48.9Gb & 250Gb of SAN

We were on 8.96.1.2 with the same config except we were on 70 users and only one jvm. We are about to move the project on to incorporate another continent and therefore 800 more users and when we started to ramp up and stress test the config it started failing. It would appear that it does not matter if we have all the users on one jvm or across 3 jvm's once we get to around 80 users the system starts misbehaving.

The Oracle database setup has been "rubber stamped" by Oracle as being able to handle easily the eventual 4000+ users we will have, so we are not too worried in that area. All performance stats show it is hardly sweating with the current load.

 

[jdel6654]

0. Run 10.1.2 Web Cache in front of 10.1.3.1

1. Oracle recommends that the Ethernet network connection be changed to Maximize data throughput for network applications.

2. Oracle recommends that the Apache HTTPD.conf setting ThreadsPerChild be set equal to 80 for quad-processor web servers. Make this change on 10.1.2 Apache Instances (httpd.conf) (WebCache instances).

3. Oracle recommends that the Apache HTTPD.conf setting UseWebCacheip be set equal to On to improve support. This setting writes the actual client IP to the logs instead of the local WebCache IP. This setting is for support purposes and not performance. Make this change on OAS 10.1.2 Apache Instances (httpd.conf).

4. Oracle recommends that the Mod_OC4J.conf setting Oc4jCacheSize be set equal to the same setting as ThreadsPerChild in the Apache HTTPD.conf file. . Make this change on both OAS 10.1.2 Apache Instances (mod_oc4j.conf).

5. In situations where an OAS installation is load balanced along equal server configurations and with WebCache in use, Oracle recommends that the Mod_OC4J.conf setting Oc4jSelectMethod roundrobin:local be added. This tells Apache to load-balance to OC4J instances on both machines but try to give priority to local OC4J instances first.

6. Oracle (JDE) recommends that the minimum memory be set to 512MB and maximum memory be set to 1024MB. This is the amount of virtual memory that the JVM will have access to. Login to the Oracle 10.1.3.1 console. Navigate to the OC4J instance and click on the link labeled Server Properties.

 

[cdawes]

We've got 300+ concurrent users on 2 horizontally clustered servers with 3 JVM's each. Servers are OAS 10.1.3.1.0 on Win2K3 Enterprise with 8 GB RAM. Backend is HP/Unix & Oracle 10g. Tools is 8.97.0.3.

The clients connect from 30+ sites across the internet (no VPN).

We are not using Web Cache, or anything more than the default compression in 8.97. Modules are all of Finance (nothing else).

Performance is more than acceptable.

The latest tuning recommendations should allow you to scale quite well.

Colin

 

[Sebastian Sajaroff]

Hi,

You should be able to accomodate at least 4 * 1 Gb JVMs
on that configuration. Which, in turn, could easily
run 300 or + users.
1. Check your JAS.INI, JDBJ.INI, JDELOG.PROPERTIES,
these parameters are critical!
2. Try not to exceed 1 Gb x JVM
3. Check your HTTP configuration (i.e. logging)
4. Check for any antivirus on the WebServer, is it
actively scanning the OAS folders?
5. Check for network bottlenecks. How many NIC cards?
Is it connected to the Enterprise via a 1 Gb or more LAN?
6. Check your JDBC drivers, are they MTR compliant?

 

[s_cambric]

When you add more than one JVM per machine or instance, does the Server Manager in 8.97 allow you to manage the users on each JVM? I know Oracle entered SAR#8363295 for 8.97 to support this capability, but wasn't sure if it was done or not?

 

[Terry Duffy]

Yes Server manager does allow you to manage users on each jvm. In that I mean you can terminate a user session / send a message to a user and block users from logging onto a particular jvm.

 

[Sanjiv Garg]

Hi Colin,

You mentioned that you have 300+ concurrent users on 2 OAS 10.3.1 servers with 3 JVM's on each. You also mentioned that you are not using WebCache. I have following questions:
What are you using for load balancing?

What are the optimised settings for JAS.INI, JDBJ.INI and JDELog.Properties for the amount of users you are running?

Do you have OHS installed on both OAS instances and using OAS clustering feature to distribute http requests to OC4J containers? If yes, which mod_oc4j load balancing algorithm is the best.

Do you use application relplication (Multicaast or Peer-Peer)?

I appreciate to answer all these questions. I am working on a project to move same amount of users from WAS to OAS.

Thanks,

Sanjiv