Re: To cache or not to cache...

[Ken Hagan <ken@weblog.org>]

My first question would be:
"Are you I/O bound?"

If I/O is not a bottleneck in your system, turning on cache will only 
help you be able to brag that you're using ATA cache.  If network connections,
processor, memory, or (perish the thought) lack of demand are your bottlenecks
then the cache won't help enough to justify any fears about corrupting an 
ext3 filesystem.

If you really want to get crazy with the power supply thing, maybe there's 
an old external device you can plug in (like an external serial modem) to 
a non-UPS outlet.  Have your system poll it and have it shutdown N minutes
after the modem disappears.  Set N to be something safe within the parameters
of your UPS and you should pretty much always have a soft shutdown when power
is the concern.  It's too complicated for my tastes but it's a way around
the UPS communication.

Just my .02USD on the issue.  

On Thu, Oct 09, 2003 at 09:18:11AM -0500, Jack Browning wrote:
> ...that is the question.
> 
> I was looking at some tuning options and ran across a thread about
> journaling file systems and the write cache on ATA drives. The gist of the
> thread was that enabling the write cache can result in an inconsistent
> (read corrupted) filesystem in the event of a power failure. Disabling the
> write cache, of course, results in a performance hit. I've got a UPS, but
> it doesn't communicate. When it runs dry and I'm not around, it's lights
> out. That sucks, but its the best I can do for right now.
> 
> I was wondering what the consensus is. FWIW, I've got two ATA drives, with
> /boot on an ext3 partition, and / and /more on reiserfs. The kernel is
> 2.4.22. Tagged command queueing for IDE drives is in 2.6 but hasn't been
> backported (AFAIK) to 2.4.
> 
> My options seem to be:
> 
> 1. Cache and hope for the best.
> 2. Turn off the cache and wait for 2.6.
> 
> TIA,
> Jack
> 
> -- 
> ======================================================
> * Jack Browning                 mojohand@charter.net *
> * 2.4.22                               1d + 17:25:05 *
> * 10/09/2003                            08:51:00 CDT *
> ======================================================
> 
> -
> To unsubscribe, send email to majordomo@silug.org with
> "unsubscribe silug-discuss" in the body.

-
To unsubscribe, send email to majordomo@silug.org with
"unsubscribe silug-discuss" in the body.