building our repository (builder)

[kraileth]

kraileth I say we go for your boss offer!

Is it possible install FreeBSD in one SSD and having a full SSD for SWAP it is just an idea? Is there any value of doing that?

Excellent! Thanks for the trust that both of you put in me. I'll tell my colleague that he can order the missing RAM and I expect that it will arrive on Monday or so. However I'm on vacation next week (leaving on Wednesday for a *nix conference hosted by GUUG (German Unix Users Group)). I'll see if I can arrange an installation before leaving, though.

Regarding swap: As far as I know it is usually preferable to distribute swap on multiple disks if possible so that more iops can be achieved. However just about any disk layout is generally possible. It is a root server and I have full access; we're not limited to any pre-configured image or something, I can do the installation. We can use a ZFS mirror of the two HDDs and use swap and UFS on SSDs or whatever we come up with. That's no big deal (unless it's so complicated that I cannot figure out how to do it using gpart and ZFS! )

[ASX]

I'll see if I can arrange an installation before leaving, though.

You will, I can tell you that you will be able to do that.

And not only that, you will install a webserver too, so that we ALL will be able to monitor the build progress using a browser; tomorrow evening you will take out your smartphone and will show your boss the build progress, so that he will know we have put the server on "good use".

[ASX]

Self note for basic reference:

http://lists.dragonflybsd.org/pipermail ... 13034.html
http://lists.dragonflybsd.org/pipermail ... 13199.html

Well, we actually dedicate two blades to doing nothing but package
building, and we can use a more substantial server if we need to get it
done faster (the record is roughly 22 hours for a full bulk build on the
dual-xeon or the 48-core opteron box)

Code: Select all

30-Nov-2016
    Initial queue size: 23476
        packages built: 23334
               ignored: 8
               skipped: 111
                failed: 23
    Duration: 22:28:12

A similar synth run that just finished today yielded:

09-Jan-2016
Initial queue size: 24701
    packages built: 24385
           ignored: 78
           skipped: 226
            failed: 12
    Duration: 20:44:20

we aim to reach half of that speed, being on a single Xeon.

[ASX]

Regarding swap: As far as I know it is usually preferable to distribute swap on multiple disks if possible so that more iops can be achieved.

While this may be true for a generic installation and a generic use of swap, in our specific case I think it is better if we use swap on a single disk: the fact is that in our specific case, most of the time the swap usage will grow depending on tmpfs usage, in turn tmpfs usage will increase due to the "build-depends" phases, which means reading packages from repository and unpacking them onto tmpfs.

Therefore I think it is better if we can have repository on one disk, and swap on a different one, and I have proposed a layout according to this view, in the other thread.

To be completely clear, what I'm looking for is to NOT have swap on the same disk as "repository", the day when we eventually would add two SSDs, we might move swap there on both SSDs. (in theory).

[ericbsd]

Yeah on that I am also thinking the SWAP shouldn't​ be on the repository disks.