CHIPP Computing Board meeting
==============================
CG_190407

Agenda for 19.4.2007 

--------------------------------------------------------------------
I. Status of current cluster (short).
    - new Hardware
    - new storage system: dCache on Solaris
    - new OS on WNs: SLC4

--------------------------------------------------------------------
II. Orientation on procurement process and current architecture proposals

--------------------------------------------------------------------
III. Decide on what to do with the old nodes (the Xeons from 2004)

--------------------------------------------------------------------

Additional information about Item III:


We need to decide on the fate of our old worker nodes at CSCS.

Some background information:
There are 15 old worker nodes.

1. The Grid information system is not able to correctly report the CPU
capacity of an inhomogeneous cluster. Agreed, this is stupid. So, one is
forced by convention to assign the kSI2k value of the slowest node to this
property. This only matters for accounting, but it is not nice.

2. The old nodes currently only share a common 1Gb/s connection to the new
dCache storage system.

3. The cluster we are using presently is the CSCS prototype Grid cluster that
they had acquired using their own funds.  We got full use of it since CSCS
had to take responsibility for the problems in the procurement of last years
upgrade. We have full usage right of this cluster until we get our phaseA
upgrade.

4. The old nodes have been drained of jobs now, so that we can upgrade them
 to SLC4.


CSCS has asked whether they could get back ~8 cores to replace a small
 cluster for meteosuisse which runs the aria project. It is completely our
 decision, because we have the full right on the new cluster. I asked them,
 whether some of our old nodes also would be ok. They said this would be 
sufficient.

So, the proposition is:

- We give 4 old nodes (=8 cores) to CSCS.

- The rest we take out of the LCG cluster, but we keep them as NorduGrid
worker nodes (because nordugrid runs the simpler jobs and I think that the
policies in respect to the information system are not as harsh). It's easy to
make a separate queue for them.

- We also could take some of the old nodes and use them as Grid User
interfaces. There people can run interactively and get acquainted with the
usage of the Tier2.

...........................................................................