We use the valgrind plugin on our ci platform to perform nightly analysis of our software. The generated results can be accessed rather easily (log files ~ 15MB/run when we don't introduce too many leaks) . But the generation of the trend graph in the end seems done from scratch everytime (we keep a history of 50 runs), which leads to huge (several minutes) loading time of the results page.

Jenkins reports that each one of these requests takes from 115 to 320 seconds, 50 of which are CPU time (which account for 96% of our total CPU usage). It's cached by the browser, so only done once a day per user, but this is still a lot when you want to have a quick overview.

I didn't check the code, but couldn't most of the graph data be pre-processed during the analysis after each build ? The other graphs that are being displayed on the same page (coverage, sloccount, test results), seem to be generated instantly, I don't know if they use caching or just need less data.