Reports 1-1 of 1 Clear search Modify search
MIF (General)
takahiro.yamamoto - 12:58 Friday 12 May 2023 (25212) Print this report
CPU load reduction of k1lsc

Abstract

I removed FilterBanks of POP_PDV1 and REFL_PDV1 which are not used yet.
Averaged CPU time of k1lsc decreased as ~5us (53us -> 48us) as shown in Fig.1.
I will check the glitch situation in a next couple of days.
  

Details

For improving the IPC glithces e.g. reported in klog#24410,
I tried to reduce FilterBank blocks from the k1lsc model.

In this time, I removed totally 10 FilterBank blocks of POP_PDV1 and REFL_PDV1.
Because we haven't had PDs in vacuum at POP and REFL yet, they makes wasted CPU load.

After removing these blocks, averaged CPU time decreased as ~5us as shown in Fig.1.
As a results, a total number of FilterBank blocks in k1lsc changed from 185 to 175.
Though removed FilterBanks were not used, de-whitening filters and gains were set.
So they seemed to have un-negligible CPU load.

Because we have a chance to check the glitch situation only once per day (around noon),
I'll check it in the next couple of days.
 

Images attached to this report
Comments to this report:
takahiro.yamamoto - 12:40 Sunday 14 May 2023 (25228) Print this report
Glitch situation seems to be changed after this model modification.
But it hasn't been solved yet...

Only 2 events were detected on Saturday (attachment 1).
But for today, 64 events were detected until noon (attachment 2).
Today's glitchy time seems to be finished.

In previous investigation e.g. klog#24949, we constantly detected ~30 events per day.
So the situation seems to be changed. But total number of glitches may not be changed as ~60 events per 2-days.

Because the duration of glitchy time didn't changed from ~1hour,
the total dead time was reduced from 1hour per day to 1hour per 2-days.
Non-image files attached to this comment
Search Help
×

Warning

×