[doc]PCI in Placement tuning
Related-Bug: #2070257 Related-Bug: #2126751 Change-Id: I602c78e02d1821fab202a91fc4e42e1b0cae42ad Signed-off-by: Balazs Gibizer <gibi@redhat.com>
This commit is contained in:
@@ -393,6 +393,15 @@ PCI tracking in Placement
|
||||
Please note that once it is enabled on a given compute host
|
||||
**it cannot be disabled there any more**.
|
||||
|
||||
.. warning::
|
||||
If you configure more than 8 PCI devices per compute with the same
|
||||
product_id or resource_class then placement needs configuration tuning
|
||||
to keep the scheduling speed acceptable. Please read the following
|
||||
two release notes:
|
||||
|
||||
* https://github.com/openstack/placement/blob/master/releasenotes/notes/bug-2070257-allocation-candidates-generation-limit-and-strategy.yaml-e73796898163fb55.yaml
|
||||
* https://github.com/openstack/placement/blob/master/releasenotes/notes/bug-2126751-optimize_for_wide_provider_trees-b50b7813fd0d9dd2.yaml
|
||||
|
||||
Since nova 26.0.0 (Zed) PCI passthrough device inventories are tracked in
|
||||
Placement. If a PCI device exists on the hypervisor and
|
||||
matches one of the device specifications configured via
|
||||
|
Reference in New Issue
Block a user