Re: [Nbd] question about i/o scheduling

[Wouter Verhelst <w@...112...>]

On Tue, Aug 03, 2010 at 08:04:29PM +0200, Peter Simons wrote:
> Hi guys,
> 
> first of all, thanks a lot for the effort you're putting into nbd. The
> package has handled my off-site backups gracefully for 2+ years, so I am
> one happy user!
> 
> Now, it looks to me like my kernel performs too much write-buffering.
> When a large file is copied to an nbd-mounted partition, the initial 2
> MB or so are transferred at incredible speed; apparently, because the
> kernel just buffers the data. At one point, the kernel actually tries to
> transfer the data to the remote host, but that takes several minutes.
> During that time, the nbd-device is almost unusable for other processes
> because it accepts no more writes at all (and reads have a hard time
> getting through the network connection).
> 
> Has anyone else observed that phenomenon? Is there maybe a way to
> improve the situation?

You can try to change the I/O scheduler. The default for most devices is
"cfq", but that isn't very useful for nbd. The 'deadline' one could
yield better results.

-- 
The biometric identification system at the gates of the CIA headquarters
works because there's a guard with a large gun making sure no one is
trying to fool the system.
  http://www.schneier.com/blog/archives/2009/01/biometrics.html