How to reduce the number of model types

[MARUBUN SUPPORT]

Hi everyone.

Our customer has the following questions.

I would appreciate some advice on how to solve this problem.

 

[Product]

Spectrum 23.3

RHEL 7.2

 

Our customer is using Spectrum while upgrading from version 10.2.1 -> 10.4.1 -> 21.2.8 -> 22.2.6 -> 23.3.

They found that the memory usage of their computer was increasing, and after investigating, they found that one of the causes was that the number of model types was over 1 million.

However, as the number of hosts they currently have registered is less than 10,000 (approximately 2,000 to 3,000 hosts), they believe that the number of model types is too high.

As they are using Spectrum while upgrading the version, they suspect that there are unused models remaining in the SSDB.

Based on the above, I would like your advice on the following questions.

 

Q1:

They suspect that SSDB has a cache in Berkeley DB and that old data may be remaining in the cache.

Please let me know if this is correct.

 

Q2:

If there is a possibility that old data remains in the cache, please let me know if it is possible to reduce the model types by clearing the cache.

Also, if it is possible to clear the cache, please let me know if there is a way to perform this maintenance.

 

Q3:

Please let me know if there are any other ways to reduce model types.

 

Best Regards,

Marubun Support

[Catalin Farcasanu]

Q1: That is not correct. 

Q2: There is no cache on the SSdb. There are only models and catalog. Some might be used, some not. There were problems with monitoring in previous versions that would increase the size of the SSdb and the number of models that exist on the system. I would do an analysis of the models that reside on the system (get a list of all model types and see their type and count) and then, with the results go for support, to see if there are problems.

Q3: no, the number of model types cannot be reduced. More model types means more specific monitoring for various types of devices. More is better.

[Gregory Polenta]

The BerkeleyDB used by the SpectroSERVER does have a cache and is set internally by the SpectroSERVER and can be adjusted by adding an entry in the .vnmrc file. This cache increased somehwere in the 21.x time frame but would not be a factor in significant memory usage. This cache is rebuilt on each SS restart.

You could use the database_tally.sh sample script to see model counts

cd $SPECROOT/vnmsh/sample_scripts/
./database_tally.sh <vnmName>

[Olaf Schalenberg]

Could it be that the 32-bit SpectroServer version was previously in use (10.2), which was limited to approximately 3 GB of memory, and now the 64-bit SS is in use, which can allocate significantly more memory?
How much memory does the SS use? 5-15 GB is quite normal in current versions.

To check for unnecessary models, you can search for All Models using the Locator and sort them by ModelType.

If you run out of model handles, consider switching to Huge Landscape Handle.