cleaner cpu usage

cleaner cpu usage

[David Arendt]

Hi,

I am currently testing nilfs on a 1TB partition with 600GB of data. I
noticed that I sometimes have through several hours 20% cpu usage by the
cleaner. During the 20% usage, I have 50% iowait, leading to an overall
slower system. Is this behaviour normal or is there something I can do
on my part. If this is normal, I think it should be possible that I
mount without cleaner and start the cleaner only if I am short on space.
Is there any argument against doing this ?

Concerning stablility, I didn't have a single crash until now.

Thanks in advance
Bye,
David

Re: cleaner cpu usage

[Ryusuke Konishi]

Hi,

On Mon, 10 Nov 2008 07:32:41 +0100, David Arendt wrote:
> Hi,
> 
> I am currently testing nilfs on a 1TB partition with 600GB of data. I
> noticed that I sometimes have through several hours 20% cpu usage by the
> cleaner. During the 20% usage, I have 50% iowait, leading to an overall
> slower system. Is this behaviour normal or is there something I can do
> on my part.

I've tried to reproduce the problem with my 1TB drive, but I didn't
see such high load for the cleanerd so far.  In my system, the cpu
usage by the cleanerd was about ~5% or so.  The phenomenon may depend
on the machine environment.  If you notice the condition that seems to
make the load higher, please let us know.

> If this is normal, I think it should be possible that I
> mount without cleaner and start the cleaner only if I am short on space.
> Is there any argument against doing this ?

Yes, you can manually stop or start the cleanerd.

Or you can possibly reduce the load by changing the following cleaner
parameters in /etc/nilfs_cleanerd.conf:

1) nsegments_per_clean (default: 2 segments)
   The maximum number of segments cleaned at a time.
   Decreasing this value slowdowns GC.

2) cleaning_interval (default: 5 seconds)
   Cleaning interval in seconds.
   Increasing this value moderates the GC frequency.


With regards,
Ryusuke Konishi