NUMA and Mosix [LWN.net]

NUMA and Mosix

Posted Jul 22, 2003 15:00 UTC (Tue) by hazelsct (guest, #3659)
Parent article: 2003 Kernel Summit: NUMA management

Mosix, or Mosix-like clustering with "shared" memory and process migration, would seem to be a special case of NUMA.

Are their plans to use pluggable transport layers (e.g. TCP/IP, Infiniband, etc.) in the Linux NUMA architecture, to bring these scalability algorithms to next-generation "Mosix" based on NUMA? It would seem this would simplify application development if the same process affinity user space calls could be used, rather than having one set for NUMA and another for Mosix.

(Log in to post comments)

Mosix and the Linux kernel

Posted Jul 22, 2003 16:05 UTC (Tue) by StevenCole (guest, #3068) [Link]

From this Linux Journal Interview with Dr. Moshe Bar:

LJ: openMosix is tightly integrated with Linux. In fact, the benefit each other quite a bit. Is there any plan to merge openMosix into official Linux kernel tree? How about porting to another platform, such as *BSD, Mac or maybe Windows?
MB: I don't want to merge openMosix into the kernel. I have talked about it with Linus, Alan Cox, Ingo Molnar and others, and I feel it is best to keep the two separate. I would love to port openMosix or part of its functionality to Windows. But I will wait for Microsoft to approach us with a proposal.

It's too bad that the interviewer didn't follow up and ask why Dr. Bar was taking that position. Sure, he said it was best to keep the two separate, but that's the same thing as not merging. The real question is why not? Does he not want to scare off Microsoft by having it merged into the Linux kernel?

Mosix and the Linux kernel

Posted Jul 29, 2003 12:19 UTC (Tue) by daniel (subscriber, #3181) [Link]

"MB: I don't want to merge openMosix into the kernel. I have talked about it with Linus, Alan Cox, Ingo Molnar and others, and I feel it is best to keep the two separate. I would love to port openMosix or part of its functionality to Windows. But I will wait for Microsoft to approach us with a proposal."

That is the nice thing about open source: Moshe Bar or anyone else cannot unilaterally keep this work out of mainline. (And incidently, I do not see how a Windows port would in any way benefit the open source community.)

Clustering is going to take off over the next few years, and even this year we're going to see the rise of home clusters. That means lots of new people will be using Mosix. When the time is right, it will go into mainline by popular demand. IMHO, the time is not yet right, it would be better to wait a while for more widespread usage, to allow the API to mature.