On Sun, Apr 24, 2016 at 10:35 AM, Rich Pieri <[email protected]> wrote: > nb: I received this Sunday morning. Not sure where the delay was. > > On 4/21/2016 10:14 AM, Jack Coats wrote: >> In many ways, we do have a single system signon, with a 'single system >> image' for well developed systems today. > > That's not single system image, though, even if it presents the > illusion. Lemme give you a concrete example, one of the last serious > attempts at practical, commercial SSI that I'm aware of: Digital's > Galaxy architecture. > > Galaxy was an evolution of VAX/VMS clustering. The idea was to be able > to construct arbitrarily large VAX clusters from what they called quad > building blocks or QBBs. Each QBB was 4 Alpha CPUs and 2GB RAM (IIRC) > and was plugged into a fast fibrecchannel backplane. The prototype I saw > in Nashua was a 4 QBB system. > > Galaxy had two configurations. The first was a traditional VAX cluster > with each QBB operating as a single node, so the prototype was a 4 node > cluster with 4 processors and 2GB RAM each. Traditional VAX clusters are > sometimes seen as partial SSI due to the cluster-aware file system (one > of the first ever) and live process migration capabilities (again one of > the first ever).
This is in fact what I think of when I hear the term SSI. In the Unix/Linux world, there was Locus (and IBM's AIX for 370 & PS/2s). You could even have both 370s and PS/2s in the same SSI. Obviously processes couldn't migrate between them, but otherwise they were treated the same. I worked with this system a bit many years ago. There is also Mosix which I believe may still be around and had a Linux based version. No experience with this > The second was the namesake galaxy configuration. All of the QBBs were > glommed together into a single large computer. In this configuration, > processes on the prototype saw 1 processor with 16 streams and 8 GB RAM > rather than a 4-node cluster. A processes could allocate more than one > QBB's worth of RAM and not hit swap. It could have threads running on > any or all of the QBBs without having to be programmed specifically for > it. Galaxy was full SSI. So memory was shared? between the QBBs? This sounds more like a NUMA architecture environment. What would you say are the differences between this definition of SSI and NUMA? Bill Bogstad _______________________________________________ Discuss mailing list [email protected] http://lists.blu.org/mailman/listinfo/discuss
