OpenSSI provides a method for a group of systems to appear as a single logical system to an application or a set of users. The seams between the systems (in terms of devices, filesystems, processes, etc) are hidden under a abstraction layer, and can be treated as a single system image (SSI). I've been experimenting with this over the holdiays, and have some progress to report.
So far, I have a partially working OpenSSI kernel on Debian (the wonders of building from source...). Process migration (halting a process on one node, moving it to another, and resuming it) is still buggy, but mostly functional. Network booting of new nodes is still not working, and I'm still chasing some problems in network address takeover (tough on a non-ARP medium). Lustre, CFS, and DLM are functional. I think we can avoid some of the network IPL issues by using IPL from a NSS instead, but will report later. If you haven't looked at this stuff, check out http://www.openssi.org. I'd really like to get this working, as it'd be a big win for having Linux as a companion to z/OS sysplexen. Now we need really high performance layer 2 networking between machines -- if we're going to migrate processes across machine boundaries, we need some really good pipes to keep performance reasonable. I *really* want multi-gigabit Etherchannel bridging capability or use of a CF interlink. IBM, how about documenting the CF, please please please? -- db David Boyes Sine Nomine Associates
