c1e9ed6d4c
The configuration is generated based on existing host's NUMA topology (as seen via sysfs) instead of a total number of cpus as it was done before. New logic attempts to load balance VMs and their cpus based on nvme drives' NUMA location. If there is no enough cpus left under the target node, all remaining nodes are checked. For the sake of the performance, cpus are not mixed between different numa nodes. Disk map is created by mapping VMs to existing nvme drives based on their NUMA location. Extra VMs are assigned in bus order of the nvme drives. SPDK cpus are split by matching the VM-to-nvme NUMA ratio. Static list can be defined as well to override this behavior. https://trello.com/c/HSoRtQkO/401-deprecate-vhost-performance-python-helper-script Signed-off-by: Michal Berger <michal.berger@intel.com> Change-Id: Ia63c6f9a472a685d252efd110eaba7b114a87d2c Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/12401 Tested-by: SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com> Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Karol Latecki <karol.latecki@intel.com> |
||
|---|---|---|
| .. | ||
| conf-generator | ||
| run_vhost_test.sh | ||