| 1 |
On 05-03-2018 18:45:00 +0100, Michael Weiser wrote: |
| 2 |
> > > gemato is down to 24 seconds for verification, hashverify now takes 13 |
| 3 |
> > > seconds. |
| 4 |
> > Darn, and how many cpu cores does it use throughout? Just two? |
| 5 |
> |
| 6 |
> I think you can relax. :) After upgrade from 10.3 to 11.2 it now runs |
| 7 |
> more than a minute again. In four runs I've seen 66, 87, 56 and 72 |
| 8 |
> seconds. |
| 9 |
> |
| 10 |
> From the PIDs, process tree and CPU load it seems to start one |
| 11 |
> controller and four workers (the i7-5650U has Hyperthreading). Work |
| 12 |
> doesn't seem to be distributed equally between the workers because they |
| 13 |
> finish at different times. The first finishes after about twenty |
| 14 |
> seconds. The other follow in 10 second intervals. At the end the |
| 15 |
> controller seems to do some more work for another 10 seconds. |
| 16 |
> |
| 17 |
> It gets slower with just two workers (-j2). For some reason I can't seem |
| 18 |
> to saturate the CPUs. It can't be I/O per se because right afterwards |
| 19 |
> hashverify goes up to 272% CPU usage and finishes in 8 to 11 seconds. So |
| 20 |
> the fs cache must be fully primed. |
| 21 |
|
| 22 |
Hmmm, I fear it's IO-bound somehow. Thanks for your detailed |
| 23 |
observations. |
| 24 |
|
| 25 |
> > > I'll go on and try to integrate gemato and hashverify with emaint sync |
| 26 |
> > > in prefix. |
| 27 |
> > Gemato should be already pretty much hooked, isn't it? I assume for now |
| 28 |
> > we could use a hook? |
| 29 |
> |
| 30 |
> It is in mainline portage. There it now lives as a hardcoded call to |
| 31 |
> gemato in rsync.py as seen with 2.3.22 on Linux. Prefix portage 2.3.18 |
| 32 |
> seems to lack it. |
| 33 |
> |
| 34 |
> Is it a lot of work to make a prefix portage release in line with |
| 35 |
> mainline portage? |
| 36 |
|
| 37 |
It's usually not that bad, but because they hardwired gemato, I don't |
| 38 |
want to push that out. |
| 39 |
|
| 40 |
> I've installed the repos.postsync.d hook from the source distribution |
| 41 |
> for now and it seems to work fine. Same for hashverify. So I now have |
| 42 |
> extra-pranoia double-verification using two implementations. :) |
| 43 |
> |
| 44 |
> How are chances to get the Prefix signing key into gentoo-keys? |
| 45 |
|
| 46 |
Good suggestion, I should see to that. Perhaps they're willing to do |
| 47 |
it. |
| 48 |
|
| 49 |
Thanks, |
| 50 |
Fabian |
| 51 |
|
| 52 |
-- |
| 53 |
Fabian Groffen |
| 54 |
Gentoo on a different level |
Archives