All, I've had a go at improving the hash thawing, and while it's not perfect (it fails on certain cases), I've managed to run november with the changes.
Unfortunately since the large number of thaws happened early in the run (when the PMCs in the packfiles are extracted) there aren't many objects to consider, and it seems to have only shaved off 5% or so (not the 20 or so I was hoping for). Will see if I can improve things over the coming days. Nick On Mon, Sep 29, 2008 at 5:28 PM, Andrew Whitworth <[EMAIL PROTECTED]> wrote: > On Mon, Sep 29, 2008 at 12:15 PM, Nick Glencross > <[EMAIL PROTECTED]> wrote: >> * Most of the GCs are happening in run_thaw to overcome a problem >> with hashes (1938 of the 2024 calls). I tried disabling the forced >> check, but it segfaults indicating that the problem is still there. >> Fixing this GC bug would allow us to shave off most of the GC time > > For the record, I ran into problems at this same exact location while > working on my GC. I couldn't do much with it because I didnt well > understand the thaw/freeze code at the time. There is definitely a > problem in this code that needs to be dealt with, if anybody has good > information about it I can take a stab. Any code that becomes less > stable without running the current GC must be in pretty bad shape. > > --Andrew Whitworth >
