You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In replay mode, RootEmbeddedFileSequence tends to do a lot of TFile thrashing opening PU files because thread assignments don't match, so a sequence that followed one thread/TFile in the previous step gets randomly assigned to threads in the following step. Caching the LFN-hash to LFN mapping would help some, caching the LFN-hash to TFile* would help a lot more.
I commented on this when I made the MixingModule threaded, but at the time it didn't seem like a problem. With more threads it looks like it is becoming more of an issue, and could be high impact for some specific WFs, especially HI production.
The text was updated successfully, but these errors were encountered:
In replay mode, RootEmbeddedFileSequence tends to do a lot of TFile thrashing opening PU files because thread assignments don't match, so a sequence that followed one thread/TFile in the previous step gets randomly assigned to threads in the following step. Caching the LFN-hash to LFN mapping would help some, caching the LFN-hash to TFile* would help a lot more.
I commented on this when I made the MixingModule threaded, but at the time it didn't seem like a problem. With more threads it looks like it is becoming more of an issue, and could be high impact for some specific WFs, especially HI production.
The text was updated successfully, but these errors were encountered: