Messages in this thread | | | Subject | Re: MOSIX and kernel mods. | From | David Wragg <> | Date | 04 Mar 1999 23:51:28 +0000 |
| |
Richard Gooch <rgooch@atnf.csiro.au> writes: > Going hopefully more on-topic, I don't think MOSIX matters anyhow. I'm > not convinced of the merit of the idea in the first place. If you want > "nice" clustering support, you're probably better off with DIPC > anyway, keeping more of it in user space (but still not all).
There is a great book about clusters, which I recommend to anyone even vaguely interested in the topic:
In Search of Clusters, Second Edition Gregory F. Pfister Prentice Hall PTR ISBN: 0-13-899709-8
It's fairly long, and assumes very little of the reader, so many readers of this list will find some parts very basic (e.g. explaining what an SMP is), but it's a really good read. It covers both software and hardware in equal measures. The 2nd ed. was revised in 1997, so the hardware details are all reasonably current. (The first edition was pre- commericial NUMA systems, but predicted them as inevitable).
In relation to this thread, it has a chapter on where you should put the layer of software that makes the cluster look like a single system. And the conclusion, unsurprisingly, is that it depends on what aspects of a single system image are important to the application, i.e. what you want to do with your cluster. The pros and cons of the various approaches are described.
It also has lots of stuff on programming models for clusters, including distributed virtual memory.
There's lots of things in there that it would be nice to see covered by free software and Linux. Beowulf clusters are fine, but technical computing is only one application of clusters, and other applications require very different techniques. Recent events such as the MOSIX episode, the announcement of a commercial clustered web-server on Linux from Pacific HiTech, and hearing that Windows 2000 will extend to 4-way shared-disk clustering have set me thinking about Linux and clusters again; it would be a pity to see free software missing in this area despite the successes of Linux.
Dave Wragg
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |