Present: Johnny Wong, Ken Salem, Ashraf Aboulnagas (prime motivators), MikePatterson, LawrenceFolland (CSCF support)

Discussions:

1. Logical configuration - network, addressing
2. Operating system to install
3. Hardware configuration - how to divide up the disks

Ashraf is quite pleased with the way the CFI 2004 clusters are set up, he proposed a head node / worker node type setup, which seemed acceptable. Ken suggested breaking up the workers into two sets, stable and unstable. The stable set would have login IDs shared with the head node (LDAP? NIS?) and nodes within the unstable set would have their own. We will give root access to CERAS members on the stable worker nodes (and unstable, of course) but not on the head node.

We're not entirely sure how networking works with the Blade Centre, there's definitely not one physical network interface per node. (Maybe it works a bit like a bridged interface in vmware terminology?)

Ken pointed out and Johnny concurred that however we set things up, it will (eventually) need to be shareable with unspecified sets of external people for unspecified purposes.

Operating system wise, the stable set should be homogenous, where unstable can differ. ("Even Windows, if that's what people want.")

Direct console access is a requirement - that should be doable.

We'll almost certainly want to RAID the disks somehow. Ken suggests allocating 5 drives for now, some sort of RAID and have the head node mount that, then re-export via NFS or whatever other sort of network-based filesharing is most convenient. Worker nodes can mount that share.

OS setup - Ashraf is happy with SuSE, nobody else objects, but we may be constrained by usage of DB2 or Tivoli. So stable nodes will run SuSE for now. We want Xen capability on these nodes as well.

-- MikePatterson - 04 Jul 2007