From: Chuck Yerkes (chuck+obsd2003.snew.com)
Date: Wed Jul 02 2003 - 01:14:30 CDT

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Quoting Sancho2k.net Lists (listssancho2k.net):
> I was hoping to get a group concensus on the use of softupdates.
>
> The FAQ says that "Soft Updates are still in development as a whole" but
> they do sound mighty appealing.
>
> Are softupdates used my many without a great danger relating to the
> problems discussed? Can I consider it safe to implement them on
> filesystems used heavily with a lot of disk activity to generally speed
> the system up?

You can read a lot on Kirk's site: http://www.mckusick.com/softdep/

At this point, FreeBSD 5.x and netbsd-current (just getting) have
FFS-2 in them which is beyond softupdates. I assume that the
adoption of it on netbsd will address some of the portability issues
that might be there (as they are watching the propolice and R^W
stuff that the Crew here have done).

I didn't use softupdates on my production boxes while it was
new. I STARTED using softupdates over 2 years ago. It's no
longer new.

It's been solid, it is faster for quick read/write files and directories.
It holds off updates of metadata - and agreggates the writes.

Kirk's an academic type, read the papers.

That said, IDE tends to be crap; cheap, but mediocre; I used
them of home systems with no qualms, however. I *do* fear
some of the disks that have done write caching without telling
the OS (if the OS says write to disk, it better not be in volatile
storage). In general, SCSI typically offers quicker performance
with multiple users (or apps). And I can get better and faster
SCSI disks. You won't likely notice so much with one user or light
use and you won't tell with linear things like "dd" - a useless
disk benchmark.

> One area that I wanted to try to improve (without forking out $$$ for
> scsi host adaptor and disks) was the performance of the system during
> anoncvs updates. My vmstat output looks like this during it:
>
> procs memory page disks faults cpu
> r b w avm fre flt re pi po fr sr wd0 cd0 in sy cs
> us sy id
> 1 7 0 111908 484404 215 0 0 0 0 0 8 0 243 522 25
> 2 1 98
> 0 7 0 111912 484400 279 0 0 0 0 0 59 0 296 569 84
> 1 1 98
> 0 7 0 111912 484400 709 0 0 0 0 0 27 0 261 821 53
> 2 2 96
>
> I understand that the 7 in the second column indicates that my disk is
> blocking and is the bottleneck. Will softupdates drastically improve
> this? (ATA 66 7200 IDE disk.)

Are you solving an actual problem or poking around.
I've found softupdates to make a difference while building
certainly. But another spindle makes more of a difference.

For a single user system, going and getting a cup of coffee and
taking a short stroll down the hall makes up for the time difference.

• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]