[haiku-development] Re: Old image

  • From: Marcus Overhagen <marcusoverhagen@xxxxxxxx>
  • To: haiku-development@xxxxxxxxxxxxx
  • Date: Wed, 15 Apr 2009 18:02:02 +0200 (CEST)

Axel Dörfler  <axeld@xxxxxxxxxxxxxxxx> wrote:

> Marcus Overhagen <marcusoverhagen@xxxxxxxx> wrote:
> > David McPaul <dlmcpaul@xxxxxxxxx> wrote:
> > > It was interesting how the machine worked while it was unzipping
> > > though.  Lots of CPU followed by a long pause (presumably IO).  
> > > While
> > > IO was going on the system was unresponsive.
> > I think this unresponsive pause is caused by the periodic execution 
> > of the 
> > synchronize cache command. This causes the hdd to flush it's whole 
> > internal
> > buffer to disk, and can take some seconds. During that time, the hdd 
> > is busy.
> > 
> > I think it is a mistake to peridically execute it. It should IMO only be 
> > done
> > once when shutting down the system.
> 
> Synchronizing the disk really shouldn't take some seconds, and does not 
> take a noticeably amount of time on any hardware I own (I really doubt 
> that's the issue the Eee is suffering from). Furthermore, executing the 
> sychronize should not freeze your system at all (only disk accesses).
> 
> Without synchronizing the disk, you would not be able to have any kind 
> of journaled file system. Since this is the status quo on all systems 
> out there, I would be surprised if any current drive would take that 
> long to write back a few blocks for real.

I have seen the described behaviour with a SATA disk, connected to Intel AHCI 
controller,
and I also increased internal timeout from 5 to 20 sec. in the ahci driver 
because of that.

Perhaps this really depends on the write cache of the hdd. Usually it should be 
off,
and in this case a cache sync shouldn't take any time and complete immediatelly.
On the other hand, with a enabled 32 MB hdd write cache full of single sector 
writes, 
flushing could take some time.

I don't think we have code that enables/disables the cache, so it depends on 
the BIOS.

Linux consider(ed?) enabling the write cache on IDE as a bad thing. Basically, 
EXT3 
doesn't seem to do a cache flush ever....


There is an interesting thread about this topic at 
http://oss.sgi.com/archives/xfs/2002-04/threads.html#01276

regards
Marcus

Zerreißen Sie die Netze der Phisher, Hacker und Betrüger!
Ihre Internet-Sicherheits-Seiten auf Arcor.de bieten alle Infos und 
Hilfsmittel, die Sie zum sicheren Surfen brauchen! Play it safe!
http://www.arcor.de/footer-sicherheit/

Other related posts: