Re: [PATCH 18/19] sched/numa: Reset scan rate whenever task moves across nodes

[Srikar Dronamraju]

* Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> [2018-06-05 10:58:43]:

> On Mon, Jun 04, 2018 at 03:30:27PM +0530, Srikar Dronamraju wrote:
> > Currently task scan rate is reset when numa balancer migrates the task
> > to a different node. If numa balancer initiates a swap, reset is only
> > applicable to the task that initiates the swap. Similarly no scan rate
> > reset is done if the task is migrated across nodes by traditional load
> > balancer.
> > 
> > Instead move the scan reset to the migrate_task_rq. This ensures the
> > task moved out of its preferred node, either gets back to its preferred
> > node quickly or finds a new preferred node. Doing so, would be fair to
> > all tasks migrating across nodes.
> > 
> 
> By and large you need to be very careful resetting the scan rate without
> a lot of justification and I don't think this is enough. With scan rate
> resets, there is a significant risk that system CPU overhead is
> increased to do the page table updates and handle the resulting minor
> faults. There are cases where tasks can get pulled cross-node very
> frequently and we do not want NUMA balancing scanning agressively when
> that happens.
> 

I agree with your thoughts here. I will try to see if there are other
workloads that benefit from this change. My rational for this change
being, because a workload consolidated and slowed down its scanning
shouldn't adversely affect it from coming back to its preferred node.


> -- 
> Mel Gorman
> SUSE Labs
>