Re: [patch] mm, slab: faster active and free stats

[Joonsoo Kim]

On Fri, Nov 11, 2016 at 02:30:39AM -0800, David Rientjes wrote:
> On Fri, 11 Nov 2016, Joonsoo Kim wrote:
> 
> > Hello, David.
> > 
> > Maintaining acitve/free_slab counters looks so complex. And, I think
> > that we don't need to maintain these counters for faster slabinfo.
> > Key point is to remove iterating n->slabs_partial list.
> > 
> > We can calculate active slab/object by following equation as you did in
> > this patch.
> > 
> > active_slab(n) = n->num_slab - the number of free_slab
> > active_object(n) = n->num_slab * cachep->num - n->free_objects
> > 
> > To get the number of free_slab, we need to iterate n->slabs_free list
> > but I guess it would be small enough.
> > 
> > If you don't like to iterate n->slabs_free list in slabinfo, just
> > maintaining the number of slabs_free would be enough.
> > 
> 
> Hi Joonsoo,
> 
> It's a good point, although I don't think the patch has overly complex 
> logic to keep track of slab state.
> 
> We don't prefer to do any iteration in get_slabinfo() since users can 
> read /proc/slabinfo constantly; it's better to just settle the stats when 
> slab state changes instead of repeating an expensive operation over and 
> over if someone is running slabtop(1) or /proc/slabinfo is scraped 
> regularly for stats.
> 
> That said, I imagine there are more clever ways to arrive at the same 
> answer, and you bring up a good point about maintaining a n->num_slabs and 
> n->free_slabs rather than n->active_slabs and n->free_slabs.
> 
> I don't feel strongly about either approach, but I think some improvement, 
> such as what this patch provides, is needed to prevent how expensive 
> simply reading /proc/slabinfo can be.

Hello,

Sorry for long delay.
I agree that this improvement is needed. Could you try the approach
that maintains n->num_slabs and n->free_slabs? I guess that it would be
simpler than this patch so more maintainable.

Thanks.