[Beowulf] SGI to offer Windows on clusters

David Mathog mathog at caltech.edu
Wed Jan 17 10:20:04 PST 2007


Mikael Fredriksson wrote
> 
> This article: "SGI to offer Windows on clusters" might be worth reading...
> 
http://www.computerworld.com/action/article.do?command=viewArticleBasic&articleId=9007859&source=NLT_PM&nlid=8
> 
> 
> Any comments?

Seems like it's going to cost $$$ in extra work to keep a cluster like
that running.

For starters it would need to have the equivalent of a Windows site
license on the cluster, or it would be pure hell to activate, image,
and so forth.  Security patches are also going to be interesting,
since the target market is Windows servers talking to Windows clients
one can't just firewall off the cluster and ignore it.  But do you
really want your production cluster running automatic updates?  I
think not!  There's also the general issues of managing so
many machines.  I guess you could set up sshd on them and run 
a lot of scripts, but historically I've always found that there's some
damn piece of Windows that's only accessible through a GUI, and
who wants to point and click a hundred times, once per node, to do the
same thing on every node? Finally there's the basic question of:
"how is this a cluster"?  Sure you can have N nodes splitting a load
under windows, typically by just shunting jobs around at the network
level, but in terms of working together as even a loosely coupled
whole how is that implemented?  Especially if the nodes are just running
off the shelf, single node type software.

For these reasons I suspect that running a big cluster of windows
machines isn't going to be an option unless there's a whole lot of
magic rolled into "Windows Compute Cluster".  And if they can roll
in that much magic they really should target that software differently,
as "Windows Multiple Workstation/Server Management Console".  That is
a tool I'd like to have now, not for clusters, but for groups of
workstations, which invariably all need the same software installs
and/or tweaks but which would take hours to reimage.

Regards,

David Mathog
mathog at caltech.edu
Manager, Sequence Analysis Facility, Biology Division, Caltech



More information about the Beowulf mailing list