How do I compile single NER on E1 Server?

[fbrammer]

Long story short, we have an NER that performs great in DV and PY, but poorly in PD. When we tested it on the fat client is performed well in DV, PY, but not in PD. After multiple experiements, including re-promoting, full builds, etc we had the following scenario:

Took a current package on the Fat Client that had the poor performing NER.
Logged onto the Fat Client and ran the test. It was slow.
Went to OMW and recompiled that one NER (no advance get or anything, just recompiled the code that was there)
It tested successfully.

Last night we did a full build and took the full build on the client and it's working great now, but the Server is still performing poorly.

This is a custom NER that's using custom Data Structures that's used in a custom interface.

What's driving us nuts is it works in PY (server and fat client), DV (server and fat client) and now PD (fat client), but it's still a bear where it matters: PD web.

I'd like to test compiling that one NER directly on the E1 server and see if anything changes.

How do I do that?

 

[jdelisths]

Are you making multiple calls to that NER or is it just one call?

 

[fbrammer]

Multiple, in almost every environment each call takes 6-10ms, but in production, each call takes over 200ms.

 

[jdelisths]

Does that 200ms include latency between the HTML server and the Enterprise Server or is it just the processing time on the server?

 

[fbrammer]

just the processing on the E1 server.

When we run it locally it runs at the expected speed. Originally, it ran slow on the Fat Client, but then we fixed it by re-compiling the NER locally. But then we did a promotion and update build, took the build on the FAT client and it went back to slow. We recompiled the NER and it went back to fast. We did a Full Build, client and server, after taking the package on the client it's fast now, but the server is still slow.

 

[jdelisths]

Can you check if there any difference in the generated C code between the locally generated NER and the one in the production package? You are on 9.1, so I assume you are using the package build process where the NER is generated only on the deployment server.

 

[Kim Schmidt]

If your NER is accessing a custom table, another possibility is that your table row counts are way different between the DV, PY and PD databases. If so,run an explain plan and/or index adviser in the different environments to see what differences may occur. Also, even with delivered JDE tables, if your NER is using a Select/Fetch Next process, make sure they are both set up with the same index.

 

[fbrammer]

We have crosswise tested everything and come down to there is something different in the configuration of the Production Servers. We Full Built to the PY E1 server and configured a test user to run through it and the process runs at normal speed. The Production Boxes are Windows EE and the rest are SE. At one point weeks ago there was no reported performance issue with this interface so something has changed. We just can't figure out what change that would have such an effect on just one interface.