RE: [squid-users] FreeBSD :Diskd ? (WAS: Compile Crash)

> -----Original Message-----
> From: Dave Raven [mailto:dave@reason.za.org]
> Sent: Thursday, December 20, 2001 12:01 PM
> To: Aaron Seelye; squid-users@squid-cache.org
> Subject: Re: [squid-users] FreeBSD :Diskd ? (WAS: Compile Crash)
>
>
> As I see it,
>
> Yes, soft-updates is highly recommended, but is _completely_
> different to
> async disk io.
> I have softupdates turned on on the drive, but I run it in async mode.
> The default fBSD disk will run on noasync (not meaning NO
> async) but that:
> Metadata I/O should be done synchronously, while data I/O
> should be done
> asynchronously.

If you read the paper by Kirk McKusick, you'll see that there is a definite
advantage to softupdates like I said before, drive state integrity. True,
your data isn't guaranteed to be written to the disk (otherwise, you
wouldn't have write-caching enabled), but when your disk comes back online,
it will be in a guaranteed usable state, unlike async mounting. If you read
tuning(7) it says "A number of run-time mount options exist that can help
you tune the system. The most obvious and most dangerous one is async.
Don't ever use it, it is far too dangerous."

>
> To enable syncronous mode only is actually shooting yourself
> in the foot,
> (I would think)
> I can only see this being done for data safety, which is
> again pointless.
> The
> chance of you loosing data is about equal.

Yes, as I said above, the chance of data loss is about equal, but you're
still able to rebuild the swap.state file, which isn't necessarily true with
async. I understand that losing your cache is the worst thing in the world,
but it's still nice to keep it around.

>
> Soft-updates has nothing to do with this, its more of a
> journaling method.
> It isnt actually journaling as it should be, but one has to
> weigh up the
> pros/cons
> of the soft-updates method.

They've never claimed it to be a journalling method. On the contrary, they
claim it to be a journalling alternative.

>
> There are a few articles about this on freebsd.org; I have no
> links though.
>
> Feel free to comment anyone, as I feel there is a general
> lack of knowledge
> as to the correct file system and what to enable / disable.
>
> (I speak for myself - as I know very little of it)
>
> Dave.
> OpteqSec.

As a side note, FBSD has recently (after the 4.4-RELEASE) added DIRPREFS.
This is something that makes for a smart layout of multiple small files in
NetBSD and FreeBSD by reserving space for files when a directory is made.
<sarcasm>Not sure where that would be helpful though.</sarcasm> :) Enabling
something like this in your kernel (by cvsup'ing to 4.4-S) and rebuilding
your cache dir's would be very beneficial to your disk speed, and I'd be
surprised if it didn't allow you to really whomp your standard async
mounting speed.

Aaron
Received on Thu Dec 20 2001 - 14:27:52 MST