I remember a few times at the Orlando ASUG and even a few places here on the board, it was generally agree’d that one of the first things you should do in a new installation is to add more Adaptive Job Servers.
My question is twofold…
First, Why exactly? Wouldn’t this be dependent on how many users you have and how many jobs? Or is there something in R4 that requires multiple Adaptive Job Servers?
Second, any particular settings one should think about? Or is a simple Clone & Rename just fine?
In 4.0 both the Adaptive Job Server (AJS) and the Adaptive Processing Server (APS) run what used to be multiple servers of different types. For example, the AJS runs what used to be the Crystal Job Server, Webi Job Server, Program Job Server, etc. Especially with the APS, it’s better to create new servers that run only one or two related server processes than to have all of them rolled into a single server. Memory management is MUCH better when you split them this way.
Hmmm I started crunching some of the numbers today and those guidelines seem a bit way out there.
By default, the APS installs in one service, with a heap size of 1 GB of memory.
After breaking mine out into approx 6 servers , my memory for the APS is between 11 and 40 GB based on recommendations (I"m leaning heavily on the 11 if not smaller).
What I noticed is that even though everything is working today, on a single box in one node, with 1 GB of memory allocated to it… they recommended amount is at least 2 GB minimum for any given process. Seems extremely high IMO…
Is anyone here running BOBJ 4 on a single box, and has gone through the sizing exercise? Wondering what others have used for thier values.
I haven’t even started to look at the Job server yet!
SAP seems to have vastly understaded the memory requirements for an Edge installation.
This link is a year newer than Marks’, and references and builds on his info, as well as the writer’s experience in working with several client builds:
The thing I liked about this one was that he put all of the WebI services in 1 large APS, so that they had a pool of memory to work from.
We are running on 1 box, 16 GB RAM, and it seems to be working just fine (even when we started at 12 GB). Of course, we are not exactly overflowing with users yet.
I’ve noticed that SAP recommended also breaking out the Adapative JOB service as well as processing, but none of the articles from users seem to mention that… wondering why.
So far my configuration was very similar to Johnathan’s. I started with SAP’s recommendation and then regrouped some of thier items that I knew I was not going to have any need for (Such as the BEx and LCM stuff).
I’m up to 9 APS, with WEBI split into two (Seperated out the visualization and DSL just in case), and I’m planning on about 16 gigs total ram. I can live with that so long as I don’t need to break out the JOB server too much if at all. I’d like the whole CMS installation to stay under 28 but could go as high as 48 if needed.
EDIT: Btw I noticed some of the recommendations talk about two services I don’t see in my installation ; “Tracelog Service” and “Insite to Action”. Are those not part of the standard BOBJ installation and some sort of addon perhaps?
EDIT #2: New question: Once I split out the APS… I noticed none of the Monitoring watches seem to work properly. For each of my new APS values, the “Health State” always reports as Unknown. I’m sure there’s some little step or checkbox I’m missing, but I was hoping someone could point me in the right direction.
JPetlev…To answer your question with regards to why the watches are not working, if you go and edit the watches and look at the rules behind the watch its tied to the old APS. Once you modify that to reference to the newly created APS Monitoring application should correctly reflect the state of things.
Yes you are correct Bob, I realized this some time after I posted. To further compound the problem, some of the APSs we created weren’t working (Even though they were set up correctly, checked by SAP). The monitoring wasn’t properly catching some of the APSs and as such there was no event to add to the monitoring system. We had to recreate the APS 3 times before one finally took, we still don’t know the cause, but since it’s working we’re not touching it
In terms of the AJS (Adaptive Job Servers). I have seen little or no need to break them out into separate containers. The AJS is different than the APS in that a child processes is created per request. This means that a separate processes is generated to manage each request therefore many of the concerns with having “all of one’s eggs in one basket” go away. In addition, it is handy having all the job servers in a single container when configuring destinations or when establishing concurrent job parameters.
There are still a few reasons you might want to break them out ranging from the need for server groups to scaleability over multiple nodes, otherwise I tend to keep them together.
With that said, I have experienced some Java Heap errors for child processes in 4.0. However this can be easily remedied by adding the Xmx?g parameter to the AJS command line start up. For example, you can add Xmx4g (comma separated) to the -Javaargs extension:
"EDIT: Btw I noticed some of the recommendations talk about two services I don’t see in my installation ; “Tracelog Service” and “Insite to Action”. Are those not part of the standard BOBJ installation and some sort of addon perhaps? "
The Tracelog Services is automatically added to each APS container as of FP3. You can no longer add it manually.
The “insite to action” was an add on as of FP3 - SP5.
I know this is an older post so hopefully you get this. Your reference to adding an increase max java heap to the Adaptive Job Server is the only document I am able to find that shows where to place this setting. The question I have to you is that it is not showing the hyphen in front of the Xmx4gb and it threw an error in the log files that it is not a valid value. So when I went back to add the single hyphen in front of it, the Expected results now show two hyphens in front of it even though I only added one. If I remove the single hyphen then the Expected results show none. I was doing this in IE8, tried in Chrome and experienced the same results. Just wondering if someone who has this setting in place could show me their Command snippet so that I can confirm how theirs looks.
(BOB member since 2002-06-20)
(BOB member since 2007-04-16)