[Kestrelhpc-developers] Status 2010-12-8

[Jon A. H. <jo...@gm...>]

On Saturday I fixed our Dracut packages.

https://launchpad.net/~kestrel/+archive/kestrelhpc/+packages


I also added some documentation about the node install and configure scripts :

- http://kestrelhpc.svn.sourceforge.net/viewvc/kestrelhpc/Kestrelhpc/Src/share/node/install.d/README

- http://kestrelhpc.svn.sourceforge.net/viewvc/kestrelhpc/Kestrelhpc/Src/share/node/configure.d/README


We need to finish kestrel-control and kestrel-daemon :

- kestrel-control needs to write a wake event the fifo file
/var/lib/kestrel/rpc/fifo before trying to wake up clients.

- check kestrel-control --wake|--shutdown|--reboot works on a real cluster.

- kestrel-daemon and register.d scripts need to be fixed, to register
the nodes when we are registering them with kestrel-control
--register.

I actually have a design problem, because I use the dhcp config file
to link the mac of a node to the node image os. But this limits us to
use a node only with only a node image os. So instead we should create
cluster node groups with kestrel-control --register [node-group] and
link a node group to 1 or more node image oses.

That means that we need to rewrite these functions on kestrel-lib :

http://kestrelhpc.svn.sourceforge.net/viewvc/kestrelhpc/Kestrelhpc/Src/share/kestrel-lib

connected_nodes()
connected_images()
disconnected_nodes()
disconnected_images()
registered_nodes()
registered_images()


And finally we need to write kestrel-launch-mpi which should be trivial.


Regards,

Jon Ander.