> For the moment we decided to reduce the WFD readout byte limit to 50K for now
> (RUN13796) which is still 10K on top of the 40K limit we had since last summer.
> The main concern is the GC on-line analysis. Kate and Alec, do you have means of
> checking the buffer collection efficiency of the ERP on-line analysis? Did you
> see any drop over the last 24 hours? Please let me know.
>
> We expect to fine tune the WFD byte limit value when the final WFD fix will
> be implemented.
>
Below are the ERP GC efficiency numbers for the last 20 runs or so.
There's no clear sudden drop at the WFD change; in fact the efficiency
is on the low side for some runs before the change. It'll take a
little more work to evaluate the statistical significance...
So, this is not an emergency... but I'll keep an eye on it (especially
since in general efficiencies are lower than I last checked anyway --
anything unusual happening on VXMACB lately?)
Kate.
============================================================
run effic
13778 0.916
13779 0.933
13780 0.920
13781 0.927
13782 0.952
13783 0.969
13784 1.000
13785 1.000
13786 0.963
13787 0.942
13788 0.955
13791 0.954 <- wfd byte limit increase
13792 0.932
13793 0.939
13794 0.960
13795 1.000
13796 0.958
13797 0.943
13798 0.931
BTW Erik: Alec suggested that these efficiencies could be calculated
automatically by one of your super-monitor jobs. I don't look at the
efficiencies all the time (although I should...it's on of those
never-get-to-the-front-of-the-queue items). To calculate the
efficiencies, I compare monitor output on disk (GC buffer event
numbers processed) to GC event numbers in the disk files. So whenever
I check this, I'm limited to whatever data files happen to be sitting
around on disk at the time (unless of course I want to go scrounging
around in the tapes). It would be nice to always have the efficiency
numbers available... Since you run through all the data anyway, your
job could possibly check the ERP GC buffer numbers against the monitor
output to get efficiencies for all runs.
Anyway, this is far from being an urgent item... just if you're interested, it
might be nice. (Or maybe I could add it to your code next time I'm
on shift).
Kate.