David Lange

L1 would seem to be the more appropriate group to assign this to. > On Aug 2, 2022, at 9:19 AM, Matti Kortelainen ***@***.***> wrote: > > > assign reconstruction...

Nothing here should be io limited:) it should be compression cpu limited > On Aug 8, 2022, at 6:10 PM, Chris Jones ***@***.***> wrote: > > > Looking at the...

There are lots of stalls - eg, Here it spent 2 hours.. Begin processing the 2712th record. Run 356993, Event 2537640, LumiSection 145 on stream 0 at 08-Aug-2022 00:05:13.389 CEST...

Whoops:) Anyway, that 2 minutes does not reproduce. > On Aug 8, 2022, at 6:21 PM, Matti Kortelainen ***@***.***> wrote: > > > There are lots of stalls - eg,...

Anyway, seems this job would take O(200) minutes on an otherwise busy cern node - maybe input files need to be copied locally to improve the cpu efficiency? > On...

Thats good to know! Thanks Chris. @germanfgv - Does the tier0 have the hardware to copy inputs locally for repack? > On Aug 8, 2022, at 7:00 PM, Chris Jones...

and if the goal is jobs that run in a shorter amount of time, one can also use fewer lumis..

I guess there is some variation in Eos read performance then... I tried rerunning my job with local files and see essentially no difference (eg, 10s of seconds improvement over...

Could you elaborate? An 8 core prompt reading from Eos takes much less than the vo card local disk requested by cms so there is extra disk around on average....

Ok, so scratch is handled differently than memory. Thats unfortunate. But so how did this work in 2018? We allocated two cores for each repack? > On Aug 10, 2022,...