Re: [Xen-devel] OpenSSI, HyperSCSI, and UBD style block devices...

[Ian P. <Ian...@cl...>]

> Before I spend the next few weeks wasting time on the following, it 
> seemed best to ask these questions on the list first.
> 
> Has anyone managed to get OpenSSI nodes running in Xen domains? While 
> you can run an older version with an ancient release of UML (before the 
> SKAS speedups), there is really no way to scale a cluster across a 
> number of hosts without running
> bare-metal. (http://www.openssi.org)

To my knowledge, no-one in Cambridge has tried this, so I suspect
you're breaking new ground.

I'd expect the OpenSSI patch to apply relatively cleanly to the
Xen-patched linux. 

[grep the patch for any diffs to the arch/i386 or include/asm-i386 
directories, but I'd have thought it would be fairly
arch-independent] 
 
> Has anyone tried a HyperSCSI or iSCSI root volume yet? It would be nice 
> to store the image volumes on a SAN/NAS server and merely have a 
> computing grid running Xen domains. 
> (http://nst.dsi.a-star.edu.sg/mcsa/hyperscsi/)

I doubt anyone has tried this either. iSCSI should just
work. However, since HyperSCSI uses raw Ethernet frames, this has
issues for the 'Virtual Firewall Router' inside Xen, which is
basically an IP router with a few L2 hacks to make things like
DHCP work. 

However, if all the HyperSCSI communication is to/from domains to
outside the machine, its probably pretty easy to make this
work. (Adding proper L2 switching support wouldn't be very hard
either).

> Instead of using VD extents for VBD devices, is there any work toward 
> UML's UBD style COW devices? It would be nice to have commit checkpoints 
>   with rollbacks, as well as single backing store volumes for a number 
> of similar nodes.

We have a CoW file system under development....

Ian