Fisheye/Crucible performance for instance with 150 perforce repositories.

[Pini Reznik]

We have a large instance with about 150 perforce repositories with corresponding number of Crucible projects and lots of groups to manage permissions on all these projects and repos.

Right now performance is terrible.

We think that reducing number of repositories, projects and groups will improve performance significantly.

What do you think about this?

Can you suggest reasons for this to work/fail?

We're going to test is in next week or two and I'll bring our results here.

[Dylan Hansen]

I would reccomend following the tips on the FishEye tuning guide to improve performance:

http://confluence.atlassian.com/display/FISHEYE/Tuning+FishEye

Often we do suggest to customers with large instances to split up into multiple installations to improve performance. Please give it a try and let us know how it goes!

[EddieW]

Would be nice if Atlassian provided some of the same suggestions they do for confluence in terms of the GC settings, heap ratios, etc. It made a huge difference in our Confluence performance. The fisheye tuning page content on JVM heam is sparce, and basically equates to "more memory!"

[Ronald Horner]

Here is how we increased performance of our FECRU instance.

Our configuration has

• 280 Repositories
• 66 Peer Review projects
• 380 Users

We increased our memory allocations. FECRU only uses the basic JVM standards when running so you need to boost these. The trick lies in the -XX:MaxNewSize flag. The Eden memory has been thrashed by garbage collection. It is a very expensive operation for Fisheye to parse your repositories and do indexing; there are lots of string operations so the number of temporary objects created is enormous. Performance also increased significantly when adding in new repositories also. Nearly all our repositories are Mercurial or legacy CVS. Now we've allocated 3GB to Fisheye but it has rarely ever used anywhere close to that. Fisheye is currently reporting 1G allocated with 501M available. The VisualVM app reports 296.5M in Eden (MaxNewSize) and 682.7M in Old Gen (Xmx - MaxNewSize)

Here are our FISHEYE_OPTS set in start.sh

-Xms1024m -Xmx3072m -XX:MaxPermSize=256m -XX:MaxNewSize=1024m

This was a huge help to us and will probably have the same effect for you. Atlassian. Please update your support page to talk to MaxNewSize

[Andrew Morin]

I am running 2.7.15 as a windows service so my memory settings are in the wrapper.conf file. I have a the following settings...

# Additional Parameters for memory
wrapper.java.additional.11=-XX:MaxPermSize=256m
wrapper.java.additional.12=-Xrs

# Initial Java Heap Size (in MB)
wrapper.java.initmemory=256

# Maximum Java Heap Size (in MB)
wrapper.java.maxmemory=1024

Is the maxmemory setting the same as MaxNewSize? Our performance is horrible with way less overhead then mentioned in this thread.

[EddieW]

ONe peice of advice is to set initial and max memory to the same number. Prevents expensive memory mapping as the heap grows. MaxNewSize is not the same as Max size,. See this great overview of the heap spaces - http://sureshsvn.com/jvm.html