{5} Assigned, Active Tickets by Owner (Full Description) (2 matches)

In order to maintain and improve server performance, we need to establish a baseline for a few different performance parameters and make sure future work doesn't affect them too adversely. The benchmarking task will handle some of this, but it's more of a high-level external view. Adding profiling support will work in tandem with benchmarking so that when something negatively impacts performance, we'll have a tool that can point us in the right direction.

Some stuff we'll need to track:

• Memory pressure, object allocations/deallocations
• Expensive methods and the call graph that leads to them
• Thread deadlock and resource contention

Based on a spike we did earlier, we think that ruby-prof will be able to provide most of these things out of the box. We'll just need to integrate it with the server in such a way that it can easily be enabled/disabled without affecting the production performance of the daemon.