Andre Przywara writes ("Re: [PATCH 1 of 3 v5/leftover] libxl: enable automatic
placement of guests on NUMA nodes [and 1 more messages]"):
On 07/18/2012 01:00 PM, Ian Jackson wrote:
What is the maximum number of NUMA nodes we might expect to see on a
single system in the next five years? I would argue that 32 is too
optimistic. 128 or 256 seem like more reasonable upper bounds.
Wow, what are you talking about?
To calm this down from the AMD side:
The current Opteron NUMA architecture is limited to exactly 8 nodes.
This has ever been the case since the release of Opteron and changing
this is not trivial and will not happen in any near future. In general I
don't think we will see much bigger NUMA systems, but more cluster like
architectures.
One of these `more cluster like architectures', if it has shared
memory at all, may well end up being most easily represented in the
model supported by Xen as a NUMA host with a large number of nodes.
We need to plan for the software that we are writing today to work for
at least the next 5 years /even if we intend to replace the algorithm
in the next release/. It is very difficult to foresee what might
happen to hardware in that time.
But it's OK, we don't need to panic. We just need a safety catch
which stops this algorithm running in situations where it won't work.
Following a discussion with Dario, AIUI he plans to implement a rule
that it will bail on systems with more than 8 NUMA nodes.
That is releasable in 4.2 and ought to be satisfactory for you ?
