Long mount requests block other mount requests

Question

Hi,

after an upgrade from NW 7.2.2 to NW 7.3.4 we have problems with mount requests. We use CentricStor VTL. If the requested tape is in cache the mount request is finished in a few seconds. If VTL needs to read data from physical tape to cache first the mount request may take about 20 minutes. If such mount request is waiting for VTL other mount requests in queue are also waiting and NW does not ask VTL to mount other tapes that are wainting in the queue. All backups are thus waiting too.

Is there any way how to run mount requests simultaneously? Any silo setting or parameter of nsrlcpd, nsrmmgd or anything else?

We need to solve this issue to comlete upgrade procedure successfully because the customer cannot accept such behaviour.

Thank you all in advance for advice.

Best regards

Petr

ble1 · Answer

As far as I remember FSC in old days, there used to be a script that was used to cache volumes before. Funny enough, this happens only with FSC.

ble1 · Answer

There are several issues with FSC and also patches - the best bet is to open this with support.

pdvorak · Answer

I opened it with support but at now it seems that they have no idea how to solve it. First thing was that we needed to find working ACSLS stuff (ssi, libstlstk, ...) because binaries included in NW 7.3.4 package do not work.

CentricStor works exactly as described in documentation. The problem is that NW runs only one mount request at one time and waits for its completion.

pdvorak · Answer

You are right. There was a so-called prefetch script that cached volumes required for recovery. Once NetWorker asked for the volumes it got a response immediatelly. On the other hand if the required volumes were not cached they did not block other mount requests because in NW 7.2.x and lower NetWorker ran all silo commands SIMULTANEOUSLY.

Since NW 7.3.x NetWorker runs all silo commands SERIALLY.

The message from EMC support is that this is a standard behaviour and just a RFE is opened for the problem.

As NW 7.3.x and higher does not use internally nsrjb to mount volumes we cannot add prefetch command to the mount sequence and a request for volume that is not in CentricStor cache blocks all ather operations.

This problem must appear in any environment where a VTL which caches physical tapes is used with NW 7.3.x or higher because as EMC told us this is a standard behavior.

pdvorak · Answer

EDL from EMC does not implement tape caching. With such VTL it should work without problem.

You will find the problem with VTLs like CentricStor, FalconStor, Copan, Sun, IBM, etc. which implement tape caching

ble1 · Answer

Not really, I have seen this only with FSC. With other VTLs (including EMC) I had to work with this behavior was not an issue. Not sure about silos as my silo is still 7.2.2 based.

ble1 · Answer

EDL is based on FalconStor. Well, I obviously didn't meet any of tape caching VTLs then - I consider that my luck. Actually I did work with Sun, but they did not use tape caching there as they didn't have any physical tapes. I guess the only thing left for you is to open RFE then (but it is strange as normally all mount request seem to run in parallel and in case that one of them can't be addressed I see others addressed with my EDL - and even it does not use tape caching mechanism should be the same).

bbeckers1 · Answer

The issue is still there in NW7.4.4.x, however I'm told by Fujitsu that their Networker version has been enhanced so that mounts are now performed simultaneously. Fujitsu was unsure however which version of EMC NW would have the enhancement included, possibly NW 7.4.5 or 7.5.2?

Does anyone have the experience that the issue is solved in those versions?

NetWorker

Long mount requests block other mount requests

Was this post helpful?