[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] RE: [Xen-devel] RE: Host Numa informtion in dom0
While I am in agreement in general, my point is that we need to avoid misleading virtualization users by somehow making it clear that "pinning NUMA memory" to get performance advantages results in significant losses in flexibility. For example, it won't be intuitive to users/admins that starting guest A and then starting guest B may result in very different performance profile for A's applications than starting guest B and then starting guest A. This may be obvious for other flexibility limiters such as PCI passthrough, but I suspect the vast majority of users (at least outside of the HPC community) for the next few years are not going to accept that one chunk of memory is *that* different from another chunk of memory. > -----Original Message----- > From: Nakajima, Jun [mailto:jun.nakajima@xxxxxxxxx] > Sent: Tuesday, February 09, 2010 3:03 PM > To: Dan Magenheimer; Ian Pratt; Kamble, Nitin A; xen- > devel@xxxxxxxxxxxxxxxxxxx; Andre Przywara > Subject: RE: [Xen-devel] RE: Host Numa informtion in dom0 > > Dan Magenheimer wrote on Fri, 5 Feb 2010 at 12:33:19: > > > It would be good if the discussion includes how guest NUMA > > works with (or is exclusive of) migration/save/restore. Also, > > the discussion should include the interaction (or exclusivity > > from) the various Xen RAM utilization technologies -- tmem, > > page sharing/swapping, and PoD. Obviously it would be great > > if Xen could provide both optimal affinity/performance and optimal > > flexibility and resource utilization, but I suspect that will > > be a VERY difficult combination. > > > > I think migration/save/restore should be excluded at this point, to > keep the design/implementation simple; it's a performance/scalability > feature. > > Jun > ___ > Intel Open Source Technology Center > > > <<attachment: winmail.dat>> _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxxxxxxxx http://lists.xensource.com/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |