Re: unusable migration daemon for 2.6

On Wed, 2005-09-21 at 12:46 -0500, Evan Hisey wrote:
> Neils-
>   hehe, Your at the same point we are at, that scheduler is a pain in
> the arse. That said, the main difference we are at then is we have
> done it in C.  Looks like the main differences are ours can be run on
> both om and none om nodes. This lets it act as a remote cluster
> monitors. And we took the apporach of letting a user agent track the
> migrate process instead of the deamon.
> 
> Evan

Hi Evan,

Thanks for the reply. The idea of non-om-monitors is cool, I'll have a
look at it :)
And really, the scheduler is difficult to understand... I thought of
solving it an other way. Not let one none make a difficult descicion,
take both (home+remote) nodes and evaluate the migration themselves.
This needs two schedulers, one for requesting the migration to an agent
and one to accept/decline the migration-request. Hopefully this approach
is a little bit easier... Still working on the thougths of the real
scheduling.
As for writing in Python, I've never done some real program with the
language before, just playing and testing :) But Python seems to be
quite easy and development goes very fast. Took me about two/three days
to produce the code I've published now.

Goodluck with the c-daemon and plaese keep me up-to-date of the status,
Niels

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.1 (GNU/Linux)

iD8DBQBDNCv9UHyL+FAJ7UYRAoGEAJ42GfZsl6KkDzEZWR9K9tNB6LLSdwCbBmwr
+sYgQokgRsSCV10bx76EtZo=
=yCqX
-----END PGP SIGNATURE-----