Re: [PATCH 00/12] Cqm2: Intel Cache quality monitoring fixes

[Stephane Eranian]

On Wed, Jan 18, 2017 at 12:53 AM, Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
> On Tue, 17 Jan 2017, Shivappa Vikas wrote:
>> On Tue, 17 Jan 2017, Thomas Gleixner wrote:
>> > On Fri, 6 Jan 2017, Vikas Shivappa wrote:
>> > > - Issue(1): Inaccurate data for per package data, systemwide. Just prints
>> > > zeros or arbitrary numbers.
>> > >
>> > > Fix: Patches fix this by just throwing an error if the mode is not
>> > > supported.
>> > > The modes supported is task monitoring and cgroup monitoring.
>> > > Also the per package
>> > > data for say socket x is returned with the -C <cpu on socketx> -G cgrpy
>> > > option.
>> > > The systemwide data can be looked up by monitoring root cgroup.
>> >
>> > Fine. That just lacks any comment in the implementation. Otherwise I would
>> > not have asked the question about cpu monitoring. Though I fundamentaly
>> > hate the idea of requiring cgroups for this to work.
>> >
>> > If I just want to look at CPU X why on earth do I have to set up all that
>> > cgroup muck? Just because your main focus is cgroups?
>>
>> The upstream per cpu data is broken because its not overriding the other task
>> event RMIDs on that cpu with the cpu event RMID.
>>
>> Can be fixed by adding a percpu struct to hold the RMID thats affinitized
>> to the cpu, however then we miss all the task llc_occupancy in that - still
>> evaluating it.
>
> The point here is that CQM is closely connected to the cache allocation
> technology. After a lengthy discussion we ended up having
>
>   - per cpu CLOSID
>   - per task CLOSID
>
> where all tasks which do not have a CLOSID assigned use the CLOSID which is
> assigned to the CPU they are running on.
>
> So if I configure a system by simply partitioning the cache per cpu, which
> is the proper way to do it for HPC and RT usecases where workloads are
> partitioned on CPUs as well, then I really want to have an equaly simple
> way to monitor the occupancy for that reservation.
>
Your use case is specific to HPC and not Web workloads we run.
Jobs run in cgroups which may span all the CPUs of the machine.
CAT may be used to partition the cache. Cgroups would run inside a partition.
There may be multiple cgroups running in the same partition. I can
understand the value
of tracking occupancy per CLOSID, however that granularity is not
enough for our use case.
Inside a partition, we want to know the occupancy of each cgroup to be
able to assign blame
to the top consumer. Thus, there needs to be a way to monitor
occupancy per cgroup. I'd like to
understand how your proposal would cover this use case.

Another important aspect is that CQM measures new allocations, thus to
get total occupancy you need
to be able to monitor the thread, CPU, CLOSid or cgroup from the
beginning of execution. In the case
of a cgroup from the moment where the first thread is scheduled into
the cgroup. To do this a RMID
needs to be assigned from the beginning to the entity to be monitored.
It could be by creating a CQM event
just to cause an RMID to be assigned as discussed earlier on this
thread. And then if a perf stat is launched
later it will get the same RMID and report full occupancy. But that
requires the first event to remain alive, i.e.,
some process must keep the file descriptor open, i.e., need some
daemon or a perf stat running in the background.

There are also use cases where you want CQM without necessarily
enabling CAT, for instance, if you want
to know the cache footprint of a workload to estimate how if it could
be co-located with others.

I think any viable proposal needs to be able to support your use case
as well as the ones I described above.

> And looking at that from the CAT point of view, which is the proper way to
> do it, makes it obvious that CQM should be modeled to match CAT.
>
> So lets assume the following:
>
>    CPU 0-3     default CLOSID 0
>    CPU 4               CLOSID 1
>    CPU 5               CLOSID 2
>    CPU 6               CLOSID 3
>    CPU 7               CLOSID 3
>
>    T1                  CLOSID 4
>    T2                  CLOSID 5
>    T3                  CLOSID 6
>    T4                  CLOSID 6
>
>    All other tasks use the per cpu defaults, i.e. the CLOSID of the CPU
>    they run on.
>
> then the obvious basic monitoring requirement is to have a RMID for each
> CLOSID.
>
> So when I monitor CPU4, i.e. CLOSID 1 and T1 runs on CPU4, then I do not
> care at all about the occupancy of T1 simply because that is running on a
> seperate reservation. Trying to make that an aggregated value in the first
> place is completely wrong. If you want an aggregate, which is pretty much
> useless, then user space tools can generate it easily.
>
> The whole approach you and David have taken is to whack some desired cgroup
> functionality and whatever into CQM without rethinking the overall
> design. And that's fundamentaly broken because it does not take cache (and
> memory bandwidth) allocation into account.
>
> I seriously doubt, that the existing CQM/MBM code can be refactored in any
> useful way. As Peter Zijlstra said before: Remove the existing cruft
> completely and start with completely new design from scratch.
>
> And this new design should start from the allocation angle and then add the
> whole other muck on top so far its possible. Allocation related monitoring
> must be the primary focus, everything else is just tinkering.
>
> Thanks,
>
>         tglx
>
>
>
>
>
>
>
>