Isaku Yamahata wrote:
> Alex started new thread whose subject "How to support NUMA?"
> Let's move the discussion there.
> The patch itself looks ok to me except the comment.
>
> Probably posting your local patches would help discussions greatly.
> Some of them might be easily merged, others might need discussion.
Hi Isaku,
Sorry, I've been meaning to get back to this, but have been swamped
here. I didn't forget you. I'll try and get back to the discussion on
the mailing list too - I'm way too behind on that one too.
I don't actually have a lot of patches, the only main one I have is one
mapping through a bunch of physical pages to dom0 as they are required
for various things on SN2 (basically our SHUB chip on each node).
Cheers,
Jes
>
> On Thu, May 24, 2007 at 04:28:21PM +0200, Jes Sorensen wrote:
>> Sorry, but there's more to it than that, there are other drivers and
>> there is the whole memory management layer too.
>
> I counted the files.
> 38 header files under linux/include/asm-ia64/sn/
> 32 C files and 5 header files under linux/arch/ia64/sn/
> Probably most of them don't need to be touched or the only
> very small part of them need to be touched.
> Yes, it might be certen amount of work to paravirtualize them.
> But I don't think still that it's unrealistic.
>
>
>> We *must* support P=M,
>> it's just not an option. So lets try and fix the real problem and not
>> introduce this really broken assumption that we can treat all memory as
>> linear.
>
> Can you elaborate on the details concretely?
> Why do you think it *must*?
> Why do you think it is so difficult to paravirtualize them?
>
> It seems many things are involved.
> Can you break them down more concrete issues?
> Can you explain Altix more? e.g. bus, shub chip, ...
> You are the only person who are familiar with Altix.
> Such infomations would be greatly helpful to find good solution.
>
>
>> In theory yes, but that means taking a hypercall for every single
>> memory allocation! How do you plan to handle CPU local and node
>> local memory pools if pages are just provided to the dom at random?
>> Without proper memory management, NUMA is not supported and the system
>> will never perform. Try treating a 64 node machine as a standard SMP
>> box and watch it behave like a 386.
>
> What is wanted here is only node locality whose requirement is
> much loser than P=M's requirement as Alex described in another mail.
>
_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel
|
Home