Bug 1384845

https://blueprints.launchpad.net/tripleo/+spec/tuned-nfv-dpdk

Is it only for DPDK?

(In reply to Yariv from comment #1)
> https://blueprints.launchpad.net/tripleo/+spec/tuned-nfv-dpdk
> 
> Is it only for DPDK?

Tune-d is definitely needed to provide affinitiy and isolation for host cores/DPDK PMD(guest)cores respectively, for OVSDPDK usecase.

In case of SRIOV i would think the re-mapping of system interrupts to the Host CPU's would be needed. 

Regards
Vijay.

Working on upstream blueprint - https://blueprints.launchpad.net/tripleo/+spec/tuned-nfv-dpdk.

Review - https://review.openstack.org/#/c/411797/

Patches upstream. Owner: skramaja

tripleo-puppet-elements review, https://review.openstack.org/#/c/418348/

Can you give a bit more context as to what is actually being done here? Is this profile being enabled for *all* Compute nodes, or only those specific subsets (Dpdk, SR-IOV, in the future real-time) that require it?

(In reply to Stephen Gordon from comment #17)
> Can you give a bit more context as to what is actually being done here? Is
> this profile being enabled for *all* Compute nodes, or only those specific
> subsets (Dpdk, SR-IOV, in the future real-time) that require it?

By default, it is disabled. And it is independent of a feature. If a particular deployment needs tune-d support for a role, the corresponding role has to add the parameter and environment file. It is applied to all the nodes in the given role.

Assuming a cluster has multiple roles, like
* Normal Compute (Compute)
* Compute with DPDK (ComputeOvsDpdk)
* Compute with SR-IOV (ComputeSriov)

then, in order to enable the tune-d profile in ComputeOvsDpdk and ComputeSriov, following parameter has to be given:

  ComputeOvsDpdkTunedProfileName: "cpu-partitioning"
  ComputeSriovTunedProfileName: "cpu-partitioning"

(In reply to Saravanan KR from comment #18)
> (In reply to Stephen Gordon from comment #17)
> > Can you give a bit more context as to what is actually being done here? Is
> > this profile being enabled for *all* Compute nodes, or only those specific
> > subsets (Dpdk, SR-IOV, in the future real-time) that require it?
> 
> By default, it is disabled. And it is independent of a feature. If a
> particular deployment needs tune-d support for a role, the corresponding
> role has to add the parameter and environment file. It is applied to all the
> nodes in the given role.

I think ideally it would actually always be enabled, as even in the "normal" compute case there is a tuned profile for that (something like virtual-host IIRC).

> Assuming a cluster has multiple roles, like
> * Normal Compute (Compute)
> * Compute with DPDK (ComputeOvsDpdk)
> * Compute with SR-IOV (ComputeSriov)
> 
> then, in order to enable the tune-d profile in ComputeOvsDpdk and
> ComputeSriov, following parameter has to be given:
> 
>   ComputeOvsDpdkTunedProfileName: "cpu-partitioning"
>   ComputeSriovTunedProfileName: "cpu-partitioning"

When it comes time to do real-time we may have to perform some extra analysis on our the usage of the cpu-partitioning profile intersects with the real-time one which contains a superset of the same options as we may want to combine on the same host.

We are waiting for Centos to package tuned-profiles-cpu-partitioning to be packaged.
Once this is done we have a review pending that needs to be pushed upstream.

Karanbir's input: tuned-profiles-cpu-partitioning is added in "rt" repo. There is a possibility to include this repo file to build DIB elements with "includepkg" set only to the specific package. We need to analyze if it is possible with DIB elements configuration.

I have enabled the centos repo with includepkgs set only for tuned profile and updated the review in the upstream.

As far as I see at /usr/share/openstack-tripleo-heat-templates/extraconfig/pre_network/host_config_and_reboot.role.j2.yaml 
The role takes _HOST_CPUS_LIST_: {get_param: {{role}}HostCpusList} as parameter for tuned (isolated_cores).
we moved to HostIsolatedCoreList as parameter for isolated cores.
shouldn't it use it as parameter?

Thanks

This (In reply to Eyal Dannon from comment #27)
> As far as I see at
> /usr/share/openstack-tripleo-heat-templates/extraconfig/pre_network/
> host_config_and_reboot.role.j2.yaml 
This file has been deprecated. 

> The role takes _HOST_CPUS_LIST_: {get_param: {{role}}HostCpusList} as
> parameter for tuned (isolated_cores).
> we moved to HostIsolatedCoreList as parameter for isolated cores.
> shouldn't it use it as parameter?
> 
> Thanks
Refer https://github.com/openstack/tripleo-heat-templates/blob/master/extraconfig/pre_network/host_config_and_reboot.yaml for the updated file. 

Use the environment file for enabling it.
https://github.com/openstack/tripleo-heat-templates/blob/master/environments/host-config-and-reboot.j2.yaml

I will provide the document with update changes for enabling it, once it has been tested (there is an in-progress patch for it).

Has been verified.

Thanks,
Ziv

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHEA-2017:3462