Stash vs. Github Enterprise scaling

[Intel CHD Jira Admin]

The issues the Atlassian team faced in scaling Stash for Git (see https://summit.atlassian.com/archives/2014/massive-teams/scaling-git) must have also been faced by the Github Enterprise team. I haven't been able to find anything about how they solved the problem.

Does Atlassian have any information about this?

Any benchmarks they could share about Stash vs. Github Enterprise scaling?

[Roger Barnes]

As far as I can tell, GitHub Enterprise doesn't scale beyond 1 node. What I do know is what we've done for Stash...

Single node instances of Stash Server have comfortably served upwards of 5000 users, thanks in part to optimisations in how Stash handles requests (both web and Git hosting). We built Stash Data Center because there are organisations with tens of thousands of developers and just as importantly, build systems that need fast and highly available access to repositories. The main hardware bottleneck we identified in production workloads was available memory and CPU resources, so a clustered solution provides the additional capacity needed at that scale.

We don't have any direct comparison benchmarks available at the moment, but we're looking to publish more detailed performance information for Stash in future. Meanwhile, we have published performance information for Stash Data Center with a particular load profile in our documentation. It shows the increased throughput possible with additional nodes.

Hope that helps!

[Intel CHD Jira Admin]

I was thinking specifically about the "packfile generation" problem addressed by Stefan in the Summit presentation, not about the data center offering. I think the Stash SCM plugin is Atlassian's solution to this problem. In order to do due diligence, I was wonder about GHE's approach here. Only throttling? Some sort of caching? The datacenter solution is beyong the scope of my investigation at present... I can't find anything in the GHE admin documentation about this. I also can't find any documentation about throttling configuration options, which I find strage. -David

[JamieA]

Why do you think they have been solved by Github Enterprise? Last time I looked, GHE was shipped as a single virtual appliance, and couldn't be clustered. 

It's the reason they are marketing it as the only scalable self-hosted git solution.

[GabrielleJ]

Isn't GHE scalable already?

[J. Caldwell]

No.