AS400 Performance

I'm sure this will be a long and interesting conversation but surely some people out there have had 400 performance issues. When we have multiple UBEs submitted to the 400 and CPU % is up around 80%-90%, web clients attempting to perform things such as a Find on an app are waiting much longer than when CPU % is low. Now granted I'm not completely in the dark here and realize that more CPU means better performance. What confuses me is that IBM says that a properly tuned 400 should run in the 80% range. If this is so then why are my clients so slow? Other servers show little or now utilization. JAS objects are on the 400 and there is a GB connection between the servers. Thanks for any input or ideas you can give me.

Todd

Technical/CNC Manager
Metro Nashville Government
AS400/Intel/WAS/WPS/Create!Form/Zero Clients

You need to provide more information. What model of AS/400 are you on? How many users? How much DASD? How many processors? How much memory?

Good Day,
By no means am I an AS400 performance specialist but one might consider looking at the max CPW you machine is capable of and then look at what you are using. For example a P10 2435 processor in an 820 has only 600 cpw. Where the new 5xx series P10 has 1000 to 2400 cpw. Just food for thought. You might want to check to see if you machine was setup for batch processes vs. interactive processes. Again I am not an expert but I found all of this out by talking with my IBM business partner. Hope this is a good starter.


--
Senior Technical Consultant
17611 E. Tennessee Pl.
Aurora, CO 80017
Work 303-337-1906
Cell 303-885-1405
Fax 303-306-1800

Todd,

There is really not enough info in your post to give you a complete answer. But I will start by saying that performance is a complex animal. Regardless of how powerful your AS400 is, there is a limit to what it can process. An AS400 that is optimally tuned for EnterpriseOne will give consistent interactive performance for applications while running a consistent number and type of batch processes. Note how I am using the word "consistent" and not "acceptable" or "good". My take on performance may be cynical but I have viewed performance tuning as a way to make sure that all users, both batch and interactive are equally unhappy. (That is a bit tongue-in-cheek but it reflects that users are rarely happy. We all want subsecond performance for our 1 million row financial reports, don't we? )

The AS400 is nice in that you can prioritize your batch load and with a little work you can separate long running, resource-intensive UBE's from quick, low resource jobs. You can do this with other platforms as well but the AS400 makes the process really easy to do.

Now, after having tuned your system in this way if you are still consistently over 80% then your only solution is to upgrade your processor capacity or reduce your workload. It is the old 10 pounds in a 5 pound bag analogy. Something has got to give, either the stuff going into the bag or the bag itself.

Take a look at these links for a good place to start learning about AS400 work management. Like all IBM manuals there is a lot in there. Unfortunately despite what the IBM marketing will have you believe about the AS400 being a magic black box, tuning is still required and takes some effort.

This link points to a list of performance tuning resources
http://publib.boulder.ibm.com/iseries/v5r1/ic2924/info/rzahx/rzahxperfbooks.htm

This link points to the Work Management manual which I feel is the best place to find out about managing your batch and interactive workload.
http://publib.boulder.ibm.com/iseries/v5r1/ic2924/books/c4153063.pdf

I hope this helps.

Regards,

I sounds like you need to set up some more memory pools and adjust what jobs are running in which pools as well as adjust some run priorities and such. The system is given to you with a *BASE pool that everything runs in.

This link is to a performance guide, really for As/400 domino, but Ch 4 gives you a great overview of how to adjust your memory usage.

http://publib-b.boulder.ibm.com/Redbooks.nsf/RedbookAbstracts/sg246937.html?Open

I also remember AS/400 is purchased with a certain amount of resources dedicated to Interactive and a percentage allocated to Batch. So even if the batch is pegged (very busy) the resources for interactive should still be available.
You may be at the point of seriously reviewing the tuning of your AS/400 which will also coincide with the tuning for your OneWorld on the AS/400, these are two sets of tuning for the same goal, better performance.

Buy a nice little NT server to run JAS BSFN's on. Spend around $8,000 on a 2 CPU x 2 GB 1U Compaq.

Map the B98305V to the machine that runs UBE's and you're all done.

Sorry I have been so long to reply (data migration). We are on a partitioned 6 way 825 iSeries. The partition for E1 contains 2 processors (2300 CPW), 11GB memory, and 1TB DASD.

We have a share pool with 6GB memory allocated to it to take care of UBEs. The slow performance with UBEs show no memory issues. DASD is hanging out around 70% usage. It's CPU that we see maxxing out.

To give the first scenario that we saw noticed this..... we were running a direct JAS gen which was using about 53% CPU. When we kicked off 3 or more UBEs the CPU went to 100+ utilization. User interaction (<20 users) went to nothing. Data fetches were taking minutes instead of seconds and users could hardly work things were so slow. Once the UBEs fell off, everything went back to normal. < 2 second fetches, interactivity was back to normal, etc... Once the JAS gen was complete users could submit UBEs with no issues. As long as CPU utilization stayed down around 50% everything was fine.

Just seems to me that something is wrong. I don't think the answer is throwing more CPU at the problem. Those UBEs shouldn't max the system in my opinion so I'm trying to figure out why they are. Your help is much appreciated.

so is your JAS/Websphere implementation in the same partition as E1? If so, have you done the work to setup storage pools and separate the Websphere from E1? If you have multiple partitions you might want to separate the two physically, otherwise, create some shared storage pools and separate Websphere from E1 - they are two different kinds of work that don't work well together.

My JAS instances are actually all on Intel servers. The only thing the 400 is running is Enterprise Services with security, database, and UBEs. BSFNs run on an Intel Logic server with the exception of the ones that have to run on the 400 (Vertex, etc...)

Todd

Hi Todd,

It would be very good to have a chat with you about performance on the iSeries.

We are a 4-way 825 on LPAR running Web on the iSeries as well. We've had serious performance issues with our implementation since go live (Apr 04) and have only recently overcome some issues on the box after involving an independant performance consultancy firm and GATS (Denver) to provide some answers. We have still engaged GATS to give our implementation a review since applying all their recommendations.

LPAR iSeries, V5R2, WS4.0.5, Xe SP22_B1