Attention is currently required from: iedemam, neels, pespin.

1 comment:

Patchset:
- Patch Set #1:
  Hi all, […]
  did you do some testing / profiling under load comparing old vs. new approach? The poll / syscall load for sure is no longer a concern. If you're worried that doing everything in one timer for all the BTSs of a BSC, it might be interesting to do some probing how long it takes between function entry and function exit of the new per-second timer function.
  Something like the followign bpftrace snippet would print you a histogram of the duration between entry and exit of the function (update interval every 10s):
```
---
/* time spent in bsc_store_bts_lchan_duration */
uprobe:/usr/local/bin/osmo-bsc:bsc_store_bts_lchan_durations {
        @timer_start = nsecs;
}
uretprobe:/usr/local/bin/osmo-bsc:bsc_store_bts_lchan_durations {
        $duration = (nsecs - @timer_start)/1000000;
        @bsc_store_bts_lchan_duration= hist($duration);
}
```
```
/* report every 10s */
interval:s:10 {
        printf("========================================================== ");
        time("%Y-%m-%d %H:%M:%S\n");
        print(@bsc_store_bts_lchan_duration);
}
---
```
  (untested, as I don't have a osmo-bsc with that new function/symbol).

To view, visit change 28165. To unsubscribe, or for help writing mail filters, visit settings.