Mark Nelson
Mark Nelson
> @markhpc The two bugs in question seem as though they'd have been triggered by any read with a non-stripe aligned offset and a length longer than a stripe. It...
@rzarzynski Thanks! I'm preparing for my presentation on Thursday, but once that's done I'll try to run this through performance tests. That will give the Teuthology tests a chance to...
@rzarzynski Are you ready to do some performance comparisons with the original version?
@rzarzynski ok, no worries. We could try running my PR through teuthology and compare results?
In general I'm for this, with the caveat that at very large OSD counts it could cause certain problems to show up faster. I'd like to see us turn pglog...
> Re the first through third sentences, do you have a sense of how large pglogs are? Are talking additional tens of MB memory per OSD? Hundreds of MB? Sadly...
@rzarzynski I can approve this, but I want to make sure we get any dissenting views into the discussion before doing so. Is there anyone on core that would like...
@MaxKellermann This is exciting! I haven't looked through this PR yet. We'll still need to support sendmsg for older kernels and there are (or at least were) some legitimate security...
> I suggest switching `ceph::buffer` to such an allocator instead of tcmalloc, at least as an option; the SSH server I wrote uses this allocator for serializing SSH packets to...
@yuriw Any idea if this one passed testing? Thanks!