Re: kswapd hogging in lowmem_shrink

[Michal Hocko <mhocko@suse.cz>]

On Fri 13-03-15 19:55:27, Vaibhav Shinde wrote:
> On low memory situation, I see various shrinkers being invoked, but in
> lowmem_shrink() case, kswapd is found to be hogging for around 150msecs.
> 
> Due to this my application suffer latency issue, as the cpu was not
> released by kswapd0.
> 
> I took below traces with vmscan events, that show lowmem_shrink taking such
> long time for execution.
> 
> kswapd0-67 [003] ...1  1501.987110: mm_shrink_slab_start:
> lowmem_shrink+0x0/0x580 c0ee8e34: objects to shrink 122 gfp_flags
> GFP_KERNEL pgs_scanned 83 lru_pgs 241753 cache items 241754 delta 10
> total_scan 132
> kswapd0-67 [003] ...1  1502.020827: mm_shrink_slab_end:
> lowmem_shrink+0x0/0x580 c0ee8e34: unused scan count 122 new scan count 4
> total_scan -118 last shrinker return val 237339
> 
> Please provide inputs on the same.

I would strongly discourage from using lowmemory killer. It is broken by
design IMHO. It can spend a lot of time looping on a large machine. Why
do you use it in the first place?

-- 
Michal Hocko
SUSE Labs

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>