Thanks for the reminder.

After adjusting swappiness to 2 on each node, I see:
http://pastebin.com/tFcAj7rw

Client buffer is now 10MB.
But map tasks seem to be slower than before.

On Mon, Jan 31, 2011 at 11:54 PM, tsuna <[email protected]> wrote:

> On Mon, Jan 31, 2011 at 10:25 AM, Ted Yu <[email protected]> wrote:
> > Here is output from vmstat:
> > http://pastebin.com/VeGC5KQr
>
> This indicates that you're actively bringing in things from swap,
> which kills performance.  The columns "si" (swap in) and "so" (swap
> out) should ideally always be 0.
>
> What's printed by /sbin/sysctl vm.swappiness ?
>
> --
> Benoit "tsuna" Sigoure
> Software Engineer @ www.StumbleUpon.com
>

Reply via email to