Re: 2.6.36 io bring the system to its knees

[Wu Fengguang]

On Mon, Nov 01, 2010 at 01:09:34AM +0000, Dimitrios Apostolou wrote:
> Hello, 
> 
> On Sun, 31 Oct 2010 09:51:32 +0800, Wu Fengguang wrote:
> > It may also help to lower the dirty ratio.
> > 
> > echo 5 > /proc/sys/vm/dirty_ratio
> > 
> > Memory pressure + heavy write can easily hurt responsiveness.
> > 
> > - eats up to 20% (the default value for dirty_ratio) memory with dirty
> >   pages and hence increase the memory pressure and number of swap IO
> 
> My experience has been different with that. Wouldn't it make more sense 
> to _increase_ dirty_ratio (to 50 lets say) and at the same time decrease 
> dirty_background_ratio? That way writing to disk starts early, but the 
> related apps stall waiting for I/O only when dirty_ratio is reached.

50% dirty ratio may help before the system goes thrashing (writing
processes will be throttled less/later). However Aidar is seeing hours
of unresponsiveness with heavy IO, in this case large dirty ratio
won't help reduce the throttling any more.

Thanks,
Fengguang
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/