Not enough memory to execute map

[jmlang22]

We are running Mercator 6.7 on a Windows2000 server and started receiveing 'Not enough memory to execute map' errors. What's strange is that when we restart the event server, the maps run without problems for the first app then subsequent apps fail. It's like the event server or some other Mercator process is not releasing memory.

We have not had any problems with this until recently and have been running essentially the same setup for several years. Has anyone else experienced this?

Thanks

 

[BocaBurger]

6.7 or 6.7.1? Have you added or upgraded anything else on the server? Windows Updates?
What is the page size and count of the map with the problem?
Use Windows task manager to watch the size of the process. Send the same data through again and again and see if it increases (on a test server).
Send your error logs, launcher or compoundsystem.txt file to support.
What changes were made "essentially the same system"?



BocaBurger
<===========================||////////////////|0
The pen is mightier than the sword, but the sword hurts more!

 

[jmlang22]

6.7.1

Does anyone have any experience with Not enough memory to execute map' errors? What exactly causes these errors? Does anyone else think Mercator is flakey?

 

[BocaBurger]

6.7.1 pre-allocates memory based on page size & count. Mostly there are two causes: 1. You don't have enough memory to process the number of threads. Fix is to reduce maxthreads. 2. Memory leak: test = run the same data through the system 10 times and see of the memory consumption has increased each run. Solution: send test case to support for engineering to fix.



BocaBurger
<===========================||////////////////|0
The pen is mightier than the sword, but the sword hurts more!

 

[c5rv]

I occasionally have this problem on my Sun box. I found that executing the offending map / file using the -W switch to force the use of a file-based workspace cures the problem. However, this is at the expense of processing time.

 

[BocaBurger]

If the map uses the sink adapater, change it to file and !create.



BocaBurger
<===========================||////////////////|0
The pen is mightier than the sword, but the sword hurts more!

 

[jmlang22]

Tried adjusting maxthreads. Set the value all the way to the minimum, 10. No help.

We are already using the -W tag in the run() function.

Sink adapter?

Are there any known external applications, patches, etc that could interfere with the running of Mercator?

 

[BocaBurger]

Maxthreads can be 1 if you want it. Many times the output card is set to "sink" if it calls a RUN map, if you don't need that card's output. If the data being passed is large, it is held in memory. Changing it to file and using a file name with the transcation set to !create (don't create) will save this RAM. If you input data is 3 meg or less, work area as memory should be OK, between 3 and 6 meg, you need to test, about 5 or 6, set the work area to file. Make sure it is -WU if you share RUN maps or the map calling the RUN map is multi-threaded. You also need to check the page size and count for both the main and RUN map. If you are using large data sizes for input or output, set StreamMaxMemLimit to 10 and make sure the directory it uses is local to the server, has enough room and you have permissions to the dir.

There have bben issues where there was a buffer overflow and a map would report out of memory, even when there was plenty available. For those instances send a reproducible test case to Ascential Support and get a patch.

Using the wrong version of JAVA can mess things up, and there have been memory issues found with Oracle clients, MQSeries and other appplications. In those cases you need to contact Oracle or IBM etc.


BocaBurger
<===========================||////////////////|0
The pen is mightier than the sword, but the sword hurts more!