Status: New
Owner: ----
New issue 1129 by [email protected]: LXC: enable CPU bandwidth control
(aka capping)
https://code.google.com/p/ganeti/issues/detail?id=1129
Hi,
currently the LXC hypervisor supports only CPU pinning. The assigned
backend parameter "vcpus" is mostly ignored. Instead of manual pinning to
physical CPUs, the cpu cgroup allows bandwidth control of CPU usage[1]. The
instance's assigned VCPUs can be directly translated into CFS quota. The
following is a quick hack to make this working:
--- a/lib/hypervisor/hv_lxc.py 2015-07-07 10:35:58.000000000 +0200
+++ b/lib/hypervisor/hv_lxc.py 2015-10-06 12:21:50.000000000 +0200
@@ -578,7 +578,7 @@
out.append("lxc.mount.entry = proc proc proc nodev,noexec,nosuid 0 0")
out.append("lxc.mount.entry = sysfs sys sysfs defaults 0 0")
- # CPUs
+ # pinned CPUs
if instance.hvparams[constants.HV_CPU_MASK]:
cpu_list =
utils.ParseCpuMask(instance.hvparams[constants.HV_CPU_MASK])
cpus_in_mask = len(cpu_list)
@@ -590,6 +590,15 @@
cpus_in_mask))
out.append("lxc.cgroup.cpuset.cpus = %s" %
instance.hvparams[constants.HV_CPU_MASK])
+ # not pinned, but limited CPUs
+ else:
+ # https://www.kernel.org/doc/Documentation/scheduler/sched-bwc.tx:
+ # By using a small period here we are ensuring a consistent latency
+ # response
+ period = 50000
+ out.append("# capped to %s CPUs" % instance.beparams["vcpus"])
+ out.append("lxc.cgroup.cpu.cfs_period_us = %s" % period)
+ quota = period * instance.beparams["vcpus"]
+ out.append("lxc.cgroup.cpu.cfs_quota_us = %s" % quota)
Maybe this is useful for someone else. Thanks, Sascha.
[1] https://www.kernel.org/doc/Documentation/scheduler/sched-bwc.txt
--
You received this message because this project is configured to send all
issue notifications to this address.
You may adjust your notification preferences at:
https://code.google.com/hosting/settings
