Bug 1397165

Summary: [RFE][Intel-OSP10] Intel RDT/MBM support
Product: Red Hat OpenStack Reporter: Krish Raghuram <krishnan.raghuram>
Component: openstack-ceilometerAssignee: Mehdi ABAAKOUK <mabaakou>
Status: CLOSED CANTFIX QA Contact: Sasha Smolyak <ssmolyak>
Severity: high Docs Contact:
Priority: high    
Version: 10.0 (Newton)CC: achernet, apannu, fbaudin, jruzicka, krishnan.raghuram, mabaakou, mmagr, pchavva, pkilambi, robert.h.armstrong, robert.w.love, sclewis, srevivo, ssmolyak
Target Milestone: Upstream M1Keywords: FutureFeature, Triaged
Target Release: 14.0 (Rocky)   
Hardware: x86_64   
OS: Linux   
URL: https://review.openstack.org/#/c/341911/
Whiteboard:
Fixed In Version: openstack-ceilometer-9.0.1-0.20171205115356.6b8591c.el7ost Doc Type: No Doc Update
Doc Text:
-
Story Points: ---
Clone Of: Environment:
Last Closed: 2018-10-18 10:09:41 UTC Type: Feature Request
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Bug Depends On: 1396152, 1457533, 1457539, 1542901    
Bug Blocks: 1476900, 1501923    

Description Krish Raghuram 2016-11-21 18:35:37 UTC
Description of problem:
Memory Bandwidth Monitoring (MBM) is a key part of Intel(R) Resource Director Technology. Availability of memory and its usage is gathered from all the nodes and made available to OpenStack to make better scheduling decisions and deliver on SLAs

Version-Release number of selected component (if applicable):
Ceilometer version (7.0.0) that goes into OpenStack Newton

How reproducible:


Steps to Reproduce:
1.
2.
3.

Actual results:


Expected results:
Available memory on each node is captured and accessible in a Nova DB for scheduling decisions

Additional info:

Comment 1 Krish Raghuram 2016-11-21 18:40:43 UTC
Supported in Linux kernel v4.6 and libvirt v 1.3.5

Comment 2 Franck Baudin 2016-11-25 14:16:56 UTC
Can you confirm that the kernel part will be implemented in https://bugzilla.redhat.com/show_bug.cgi?id=1315489 ? Thanks!

Comment 3 Krish Raghuram 2016-11-29 16:25:54 UTC
(In reply to Franck Baudin from comment #2)
> Can you confirm that the kernel part will be implemented in
> https://bugzilla.redhat.com/show_bug.cgi?id=1315489 ? Thanks!

Bugzilla 1315489 is for an Intel CMT/CAT package support as an alternative (or maybe stop-gap, since it seems to be only landing in EPEL) to full support in the kernel. I'm investigating when the kernel support shows up in RHEL (7.4 or 7.5), and will provide an update soon

Comment 4 Krish Raghuram 2016-11-30 20:10:36 UTC
(In reply to Krish Raghuram from comment #3)
> (In reply to Franck Baudin from comment #2)
> > Can you confirm that the kernel part will be implemented in
> > https://bugzilla.redhat.com/show_bug.cgi?id=1315489 ? Thanks!
> 
> Bugzilla 1315489 is for an Intel CMT/CAT package support as an alternative
> (or maybe stop-gap, since it seems to be only landing in EPEL) to full
> support in the kernel. I'm investigating when the kernel support shows up in
> RHEL (7.4 or 7.5), and will provide an update soon

I've been advised RHEL 7.4/7.5 will be based on kernel 3.10+. I'm not aware of any plans to backport the kernel 4.6 addition to kernel 3.10+, so we will have to rely on the out-of-kernel implementation at https://admin.fedoraproject.org/pkgdb/package/rpms/intel-cmt-cat/ and its related bugzilla https://bugzilla.redhat.com/show_bug.cgi?id=1351707

Comment 5 Krish Raghuram 2016-12-02 16:39:43 UTC
(In reply to Krish Raghuram from comment #4)
> (In reply to Krish Raghuram from comment #3)
> > (In reply to Franck Baudin from comment #2)
> > > Can you confirm that the kernel part will be implemented in
> > > https://bugzilla.redhat.com/show_bug.cgi?id=1315489 ? Thanks!
> > 
> > Bugzilla 1315489 is for an Intel CMT/CAT package support as an alternative
> > (or maybe stop-gap, since it seems to be only landing in EPEL) to full
> > support in the kernel. I'm investigating when the kernel support shows up in
> > RHEL (7.4 or 7.5), and will provide an update soon
> 
> I've been advised RHEL 7.4/7.5 will be based on kernel 3.10+. I'm not aware
> of any plans to backport the kernel 4.6 addition to kernel 3.10+, so we will
> have to rely on the out-of-kernel implementation at
> https://admin.fedoraproject.org/pkgdb/package/rpms/intel-cmt-cat/ and its
> related bugzilla https://bugzilla.redhat.com/show_bug.cgi?id=1351707

I've been further advised (!!) that many Telco customers wont accept the out-of-kernel EPEL package option and must have kernel support. I'll be logging a bugzilla for kernel support (ie. backporting) in RHEL 7.4 and tracking that, along with support in libvirt/RHEV

Comment 6 Franck Baudin 2017-01-06 14:12:28 UTC
> I've been further advised (!!) that many Telco customers wont accept the
> out-of-kernel EPEL package option and must have kernel support. I'll be
> logging a bugzilla for kernel support (ie. backporting) in RHEL 7.4 and
> tracking that, along with support in libvirt/RHEV

Can you add the RHEL BZs (kernrl and libvirt) as a dependency for this BZ? Thanks!

Comment 7 Krish Raghuram 2017-03-16 21:05:40 UTC
(In reply to Franck Baudin from comment #6)
> > I've been further advised (!!) that many Telco customers wont accept the
> > out-of-kernel EPEL package option and must have kernel support. I'll be
> > logging a bugzilla for kernel support (ie. backporting) in RHEL 7.4 and
> > tracking that, along with support in libvirt/RHEV
> 
> Can you add the RHEL BZs (kernrl and libvirt) as a dependency for this BZ?
> Thanks!

Latest update is that the kernel upstream changes will only land in v4.11, which is too late for RHEL 7.4 backporting. We will now target RHEL 7.5 in '18 and provide the BZ# when it is lodged. Libvirt changes are all done

Comment 9 Mehdi ABAAKOUK 2017-07-12 07:39:46 UTC
The puppet/tripleo stuffs is the same as the CMT feature and already done: https://bugzilla.redhat.com/show_bug.cgi?id=1396152

Comment 10 Mehdi ABAAKOUK 2017-07-12 07:46:53 UTC
Krish as requested by Franck, can you add the RHEL BZs (kernel and libvirt) as a dependency for this BZ? 

I have found 1457539 for the kernel side, is that correct ? 

but nothing for the libvirt side.

Comment 11 Mehdi ABAAKOUK 2017-07-12 07:54:28 UTC
For libvirt, that already looks OK, rhel-7.3 have 2.0.0 and rhel-7.4 have 3.2.0.

Comment 12 Krish Raghuram 2017-07-25 15:49:07 UTC
(In reply to Mehdi ABAAKOUK from comment #11)
> For libvirt, that already looks OK, rhel-7.3 have 2.0.0 and rhel-7.4 have
> 3.2.0.

Correct, libvirt work for MBM was done as far back as 1.3.5, so it is all present in 2.0.0 and RHEL 7.3 itself(In reply to Mehdi ABAAKOUK from comment #10)
> Krish as requested by Franck, can you add the RHEL BZs (kernel and libvirt)
> as a dependency for this BZ? 
> 
> I have found 1457539 for the kernel side, is that correct ? 
> 
> but nothing for the libvirt side.

Yes, BZ# 1457539 is for RHEL 7.5 Thx for finding it!

Comment 17 Mehdi ABAAKOUK 2017-10-26 13:03:37 UTC
I have tested that on OSP12 and it already works.

It have to be enabled to works, documentation should be the same as the CMT here: https://bugzilla.redhat.com/show_bug.cgi?id=1475002

Comment 18 Mehdi ABAAKOUK 2017-10-26 13:59:38 UTC
Any idea why we need to rhel 7.5 ? the rhel 7.4 kernel seems to already support MBM.

Comment 19 Krish Raghuram 2017-10-26 15:08:28 UTC
(In reply to Mehdi ABAAKOUK from comment #18)
> Any idea why we need to rhel 7.5 ? the rhel 7.4 kernel seems to already
> support MBM.

Mehdi, I believe the kernel changes mentioned in https://bugzilla.redhat.com/show_bug.cgi?id=1457533 are needed for proper support of CMT & MBM (old perf-based method was deemed unacceptable by some in the community). Maybe that's only for some corner cases. Either way, thanks for completing the testing

Comment 20 Mehdi ABAAKOUK 2017-11-30 11:44:33 UTC
I changed the state to POST, since everything is done.

We just need to wait the first rhel75 and OSP13 puddle to be out to be able to put MODIFIED.

Comment 35 Mehdi ABAAKOUK 2018-08-29 09:34:50 UTC
cmt, mbm_total, mbm_local perf events kernel support have been removed from rhel 7.5, see:

https://bugzilla.redhat.com/show_bug.cgi?id=1542901

It seems, there is no plan to add back the support of this in the RedHat kernel.

So, I wonder what we should do we this RFE, CLOSE CANFIX ?

Comment 36 Martin Magr 2018-10-18 10:09:41 UTC
Closing according to comment #35.

Comment 37 Pavan Chavva 2018-10-18 11:52:33 UTC
Hi Maritn,

The bug 1542901 is a regression and it is expected to be fixed.

Can we keep this bug open till then?

I will follow up with Intel and post an update.

Best,
Pavan.

Comment 38 Martin Magr 2018-10-19 14:59:26 UTC
The original bug (#1532553) has been closed as CANTFIX too. Are you sure it is expected to be fixed?