Re: MultiDatabase shard count limitations
Olly Betts <[email protected]> Mon, 24 Aug 2020 05:23:45 +0100
| Newsgroups | gmane.comp.search.xapian.general |
|---|---|
| Message-ID | <[email protected]> |
On Sun, Aug 23, 2020 at 11:12:39PM +0000, Eric Wong wrote: > Olly Betts <[email protected]> wrote: > > If most (or at least a significant amount) is CPU time then it would > > be useful to profile to see if there are any low-hanging fruit. I've > > been mostly using the profiler in gperftools lately if you want to > > try this and don't know what to use. > > Haven't tried gperftools, but I'm somewhat familiar with Linux `perf`. > Just installed the -dbgsym packages from Debian, attached is > "perf record ... && perf report" output. The report you show seems to be just the time take by each function directly rather than including functions it calls. It looks like a lot of the time is spent in cursor movement, as totalling up things that seem like they'd be due to that I quickly get to 40+% but it's hard to tell if that's about the actual total or if this is really where almost all time is spent. If it were 50% then getting rid of it all would only halve the time which would be a nice speedup, but still too slow for your case (and some cursor movement is required as that's how we skip ahead in posting lists). Can prof report time for a function including things it calls? Cheers, Olly