should migrate package-import.ubuntu.com onto standard centrally-managed hardware

[Martin Pool]

At the moment the package-importer is on a really large machine that's a twin of the landscape database server. IS would like us to move this off this machine; at the same time perhaps we can make it a more standard losa service.

This requires:

 * checking whether it actually needs an unusually large machine
 * allocating a new machine (needs an rt)
 * migration
 * handing off any special privileges from james_w to losas

[Martin Pool (mbp)]

james_w says: fine cpu wise, is using quite a lot of memory, don't know about io load. does use a lot of disk but much of it can be deleted.

[Martin Pool (mbp)]

After some brief inspection with elmo of jubany: it's bursting large amounts of IO which are quite quick because it has 16 spindles, but the average rate is pretty low. Moving it would go from 32GB ram to 6GB (though with the option to add more), and that may be enough. There's 240GB of disk used, which is more than the typical server, but if some of that is cache we may be able to cope.

Unless someone else screams, elmo will pick a time to move it, giving a few days notice to ~canonical-bzr and the udd list.

[Robert Collins (lifeless)]

JFDI please!

[John A Meinel (jameinel)]

I think part of it is that the managing script always runs 8-concurrent imports (probably to match the 8-cpus). I don't know if it needs to be running that quickly, and that would probably lower the peak memory consumption, etc.

[Vincent Ladeuil (vila)]

As a data point, nexuiz-data currently hang (for yet unknown reasons) while using 4.6G.

There is obviously a bug there but this doesn't mean the 4.6G are not needed either.