[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH 0/6] xen,xend,tools: Add NUMA support to Xen

I've respun the NUMA patches against 10874 and I'm re-submitting them
with the optimizations mentioned [1]previously on the list.  There was a
request to see the overhead on non-numa/single-node machines.  I've
re-run those benchmarks (ballooning up from small mem to multi-gig) as
well as timing the initially domain increase_reservation time to gauge
the overhead when allocating from the heap.

The results for these were done on an x460, single node, with 6G ram.
The balloon test was from 512M to ~6G.  The NUMA patches include all of
the previous optimizations.

Balloon Up:
Without NUMA    With NUMA    NUMA Delta
Try1: 2943ms    2937ms       -6ms
Try2: 2996ms    2992ms       -4ms
Try3: 2971ms    2968ms       -3ms

Increase reservation
Without NUMA:
MemSize  128M  512M 1G   2G    3G    4G
Try1:    10ms  38ms 76ms 160ms 237ms 308ms
Try2:     9ms  37ms 76ms 151ms 232ms 308ms
Try3:     9ms  37ms 76ms 151ms 226ms 311ms

With NUMA:
MemSize  128M  512M 1G   2G    3G    4G
Try1:    10ms  42ms 84ms 170ms 260ms 346ms
Try2:    10ms  42ms 83ms 167ms 255ms 346ms
Try3:    10ms  42ms 83ms 168ms 251ms 346ms

The only other change has been to the infrastructure patch in which I
added the call to Xen's command line parsing to pick up the numa=
options.  This allows even on NUMA systems to disable NUMA parsing and
fake the number of nodes (numa=off fake=1, would fake 1 node).

1. http://lists.xensource.com/archives/html/xen-devel/2006-07/msg00384.html

Ryan Harper
Software Engineer; Linux Technology Center
IBM Corp., Austin, Tx
(512) 838-9253   T/L: 678-9253

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.