Does anybody know of the full steps in building a "slave" application server in Windows ? I built a slave application server (Windows 2003) based on ERPLN's guide - "Installation and Configuration guide for the Application Server" (U8392B US) to run jobs against with. This method allows me to run heavy ERPLN jobs outside of the "master" application server relieving the load off to the slave. However, I find that not all jobs can be run against the slave because the porting set installation does not create all the folders required by some sessions (log, tmp, audit, etc.). These folders has to be centrally sourced from the master server and mounted on the slave server. But Windows 2003 does not allow hard links of a network share to a local folder. Now that the Porting Set 8.5a03 has been certified by Infor to work on Windows 2008, using mklink I will be able to mount the central master folders. However, I still need full information on how to complete an installation of a slave application server.

The idea is to grow the ERP system horizontally; being able to use multiple VMs as application servers; taking advantage of Enterprise Windows licenses and dividing the ERPLN load between these slave AS. Example application would be: warehouse gets their own AS, one for Accounting, one for the job server, one for bus components, one for remote sites, etc.

Please, I need your rare information because I need to ease up the ERPLN load that is fast becoming a problem currently.

TNX.

Jojo ****ebo
Systems Administrator
PRIDE Industries

P.S. I have information on how to setup an isolated network using VMs to be able to implement anenvironment for test, training, testing feature pack, code development, etc. Email me at erpln_baanboard@lagunarty.com

Howdy,

Are you using a 3 tier setup or a 2 tier?

The app server really shouldn't be doing a whole heck of a lot, how many users are we talking about?

What most people do is 1 app server and a DB cluster if they need to scale horizontally.

If the app server is really hitting a high load, then it's not particularly difficult to scale it out. But I'd be pretty surprised to hear that it was.

I don't particularly like the method you're using for your cluster. To do that properly you need a shared disk clustering software like Veritas. But I think there is a better method - Let me think on the multi-VM setup. might be the subject of a good post on Baan in the Cloud.

Dave

I think I have a 3 tier; I have a Windows 2003 SQL2005 server, a Windows 2003 ERPLN Master Application server MAS), and most clients use a single Windows 2003 Webtop server. We have about 150 licenses and we are not implementing session timeout (I can't figure why) therefore we hit 150 ntbshell sessions on the Master AS. We see some desktop ERPLN slowdowns randomly almost twice every week. When users run a session that uses reportviewer we see a hit on the SQL and ERPLN server; this is not to say this is the cause of our slowdown.

We also have a ton of printers installed on the same MAS. The idea was to dedicate ERPLN printer shares for ERPLN only, but it seems that it is causing overload too. We are intending to move this off soon.

What do you mean of a DB cluster ? Is this an SQL cluster working in hot-standby mode ? Or load-balanced mode ?

What utility can I use to determine if there is an ERPLN latency ? All the users get disconnected when we get the slowdown, I see lots of communication link event log error and, once in a while, 10053 event log errors. What is the maximum number of ERPLN users can an appserver with 16GB RAM and 2.4GHz x 4 CPU on a VM take ? Also, when we get these errors we lose a ping to the app server and to the webtop server.

I think the shared disk clustering may not be needed as Windows 2008 can mount network shares onto a folder (mklink). Also, I've been informed by our PSO who has left the company that this is done in the Unix world and that ERPLN will handle the management open files, therefore, there is no requirement for file system clustering.

I do need to know if there really is a need for multiple app servers, otherwise, there is something wrong with our systems. If so, I advocate for scaling ERPLN horizontally but I need to acquire the correct process for this.

Thanks for your thougts.

JP

What utility can I use to determine if there is an ERPLN latency ? All the users get disconnected when we get the slowdown, I see lots of communication link event log error and, once in a while, 10053 event log errors. What is the maximum number of ERPLN users can an appserver with 16GB RAM and 2.4GHz x 4 CPU on a VM take ? Also, when we get these errors we lose a ping to the app server and to the webtop server.

hmm 150 users shouldn't require two servers and that box is plenty big. I think something else is wrong.

To me that smells of network troubles, specifically an issue with the NIC card on the app server getting flooded.

I'm not that familiar with windows, but I once had an issue where i had a performance slowdown on a windows box and we got into the driver specific parameters for our NIC interface and found there was a 'large packets' option we needed to turn on. Maybe that would help you.

Dave

We too have similar users working with substantially low config in windows cluster on Baan IV with no performance problem .

Here are certain things u can do.

1. If max memory on SQL server is set to unlimited figure , set it to 90% of physical memory . Also check what is the page file size . Does SQL server page too much . I think setting this max memory limit will certainly solve the problem.
2. Clear the BSE/tmp folders regularly.
3. Avoid disk based data transformation and use in-memory data transformation while creating report / update sessions that summarises figures . Check this board for in-memory data transformation / performance dll. This is a general suggestion for increasing the health of hard disks and application servers which also improve users speed drastically.
4. Think u use Windows server printer . I think that should not be a problem. As a last alternative u can try using windows printer.
5. MAS may have gbps card . But check the switch / cabling etc if it's capable of matching server NIC speeds.

Dave,

I want to suspect the network as a problem too. The server used to be a physical server, then we P2Ved (clone) it to a VM server. I am not sure if the NIC stack is clean after doing something such as P2V. As for a test we are running "pinging the MAS server and webtop server" and "pinging from the MAS server to a non-ERPL related server". Whenever we get an ERPLN disconnect event, the MAS and the webtop server will lose ping at about the same time (but no other VM guest server on the same VMWare host will lose a ping), and that pinging from the MAS server will not lose a ping. I am starting to believe that all I'm losing are all ERPLN packets as the ERPLN application stack becomes too much loaded. I need to figure out how to run multiple bshells and sessions against my test server and figure out the point of the ERPLN saturation while running another application on the same test server.

JP

Hitest,

1.SQL memory - We have it set at about 91% of total OS memory.
2. We clear this daily but the number of tmp files grow fast. We are not keeping tabs on it but as of 8:30 this morning there are 1000 files considering that we clean it up ever 3 AM.
3. Performance dll - this seems to be a project but I will forward the information to our in-house developer.
4. Windows printer - Users will be able to use the windows printer if they use "default". This is not to say that they are not using Windows Printer driver that may add additional load to the server. The Windows printer drivers were configured in ERPLN for desktop-less clients such as bar-code scanners, etc.
5. Server Maximum NIC speed - I will have to devise a test (possibly IOmeter) to figure out the saturation point of the server's NIC.

TNX all for the replies. I will setup these test scenarios and post again later.

JP