MII Application Performance Issues

[Former Member]

Hello Experts,

NW Version: 7.30

MII Version: 12.2.5 Build(72)

OS: Windows Server 2008 R2 Enterprise SP1

CPU: Intel Xeon X5675 @ 3.07 GHz (2 Processors) (6 Core Each)

RAM: 48 GB

LAN Speed: 100 MBPS

Our application runs on the above NW & MII version and NW is configured to use 39 GB RAM. We have a separate DB server(MS SQL 2008) with the same CPU, but, with 80 GB RAM and we use MDO tables to persist the data. We face critical server issues, where, the JAVA nodes stop and re-start multiple times and also lead to "Out Of Memory" and "Cannot Allocate Connection" exception on the Log Viewer. Response time of the application is very slow and every transaction call from the UI takes a minute to execute.

We have 400+ users on the shop floor and they make heavy use of the MII application. MDO's are used pretty heavily. In order to decrease the load on the server, we have multi-layered asynchronous transaction calls to pull & post data to ECC via RFC's. When the server is severely loaded some of the asynchronous transactions do not execute and goes to PENDING status which leads to data issues.

Based on some SCN recommendations we changed the following NW parameters on the Application Thread Manager side:

ExecutorPoolMaxSize from 60 to 120

MaxThreadCount from 300 to 600

PercentageofParallelismAllowed from 30 to 60

DB Connection Pooling Maximum Connections from 30 to 200

Initial & Max Heap Size from 4 GB to 6 GB.

In-spite of the above changes our servers are still slow and we need to constantly monitor them and re-start the SAP MII & Java services to free up the system resources. Memory Consumption is at 55-85%, CPU ranges between 3 - 60% during heavy loads. We have constant Heap Dumps, Full GC Duration takes 6 sec, Paging on the OS takes ranges between 10 - 16000 pg/sec during high loads.

Any advise is helpful.

[Former Member]

Could you resolve this?

Regards.

Federico.-

[former_member185280]

It is probably best to have a Netweaver expert from sap to take a look.

What is the FCAServerThreadCount setting? This is the amount of threads the ICM/http? dispatcher can use. The ICM passes requests back and forth beetween the back end java and the clients. So you can have a lot of back end threads but an insuficient number at the ICM causing a bottleneck.

[Former Member]

Hello Christian,

We have the default FCAServerThreadCount setting of 15. I also checked the NWA under "Availability and Performance" --> "System Overview", the ICM Request Queue is always 0. The Heap Memory gets filled up very rapidly too.

[former_member185280]

What's the thread pool usage rate? The symptoms you are describing sounds like servers I have seen that are just maxed out and can't keep up with all the thread and memory management and once it hits the tipping point of constantly trying to swap its all over. More memory might help.

It may be that you have maxed out what the server can handle regardless of any possible tuning and you might want to look into a bigger or multiple servers. Maybe there are some background processes you can unload onto a separate mule box.  Also depending on how you implemented your asynchronous process if its not throttled you may be using up all your threads and I know from experience that Netweaver really won't do anything to stop you. Get an expert from SAP to look at it.