Elena Ufimtseva
2013-Sep-13 08:50 UTC
[PATCH RFC v2 6/7] xl.cgf documentation update for vNUMA.
xl.cgf documentation update for vNUMA. Signed-off-by: Elena Ufimtseva <ufimtseva@gmail.com> --- docs/man/xl.cfg.pod.5 | 50 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 50 insertions(+) diff --git a/docs/man/xl.cfg.pod.5 b/docs/man/xl.cfg.pod.5 index 08d6cc4..0be0c7b 100644 --- a/docs/man/xl.cfg.pod.5 +++ b/docs/man/xl.cfg.pod.5 @@ -208,6 +208,56 @@ if the values of B<memory=> and B<maxmem=> differ. A "pre-ballooned" HVM guest needs a balloon driver, without a balloon driver it will crash. +=item B<vnuma_nodes=N> + +Number of N vNUMA nodes the guest will be initialized with on boot. Defining this +along with memory layouts turns on vNUMA initilization for domain. +In case numa=off kernel boot parameters or kernel was not compiled with CONFIG_NUMA, +vNUMA topology will not be constructed. + +=item B<vnuma_mem="vmem1[m|g], vmem2[m|g], ..."> + +The memory sizes should be defined to enable vNUMA along with nr_vnodes. +The node sizes should be prefixed with g (Gbytes) or m (Mbytes). In case vNUMA +nodes memory is not equal to memory, the domain memory will be adjusted to total +of vNUMA memory. +Memory allocation policy for vNUMA-enabled domains on NUMA enabled hardware +uses auotmatic NUMA allocation node map and will try to fit all vnodes on single +physical node. If node map spans multiple physical nodes, vnodes will be placed +to physical nodes one by one with effort to preserve locality. In worst case +scenario, when for some of the vnodes the physical node cannot be found due to +the memory constraints, default (sparse) allocator will be used. This will +render suboptimal performance and should be avoided. +For UMA machines, allocating domain vnodes will result in default allocation, as +no vNUMA was present. + +=item B<vdistance="d[0][0], ..., d[n][n]"> + +Defines the distance table for vNUMA nodes represented by two dimensional array +where n is number of nodes minus 1. If not defined in config, uses +default table: distance equals 10 for same node, 20 for other elements of +distance table. If vdistance was defined incorrectly, default distance table +will be used. + +=item B<vcpu_to_vnode="vnode1, vnode2, ..."> + +Defines vcpu to vnode mapping as a string of integers, representing node +numbers. If not defined, the vcpus are interleaved over the virtual nodes. +Current limitation: vNUMA nodes have to have at least one vcpu, otherwise +default vcpu_to_vnode will be used. + +Examples: + a) + vnodes = 2 + vnumamem = "512m, 512m" + b) + memory = 2048 + vcpus = 6 + vnodes = 2 + vnumamem = "1g, 1g" + vnuma_distance = "10 20, 10 20" + vcpu_to_vnode ="1, 0, 0, 0, 1, 1" + =back =head3 Event Actions -- 1.7.10.4