[linux-audio-user] is ext3 ok for real-time / low-latency?

[Larry Troxler]

On Saturday 15 February 2003 08:57, Jan \"Evil Twin\" Depner wrote:
> I have copies of Mark Knecht's benchmarks on my web page :
>
>  http://myweb.cableone.net/eviltwin69/Arcana.html

That's a good link - thanks.

Man, there's so many factors to think of it. I was especially boggled to hear 
that the disk drive cable I'm using might be a culprit. Arggh! Who can keep 
track of all this?

>
> as well as a write-up explaining why you shouldn't use ext3.  It
> basically comes down to the fact that ext3 is using a separate file to
> handle the journal.  What this means is that as you write your audio
> data, every once in a while, the system has to write to a different file
> in a separate location on the hard drive. 

Yeah, that's what I was concerned about. 

> It will depend on how close
> the files are physically to each other, disk latency, and a host of
> other things but, eventually, you will see problems with ext3.  It's not
> hard to convert to reiserfs (instructions (destructions?) are included
> on the above page) so why not. 

Looks like you can't convert non-destructively though. What a pain.

> Reiser journals are kept with (as part
> of?) the files - you don't have to run fsck after a crash.  A real
> intersting thing to note is that reiserfs actually seems to be faster
> than ext2 for what we're doing.
>

fsck'ing is a pain, yes, but if it needs to be done too often there's other 
problems :-) so for me, ext3 is nice, but not needed I think. I have had bad 
shutdowns over the years quite often on ext2, and although it's not nice to 
wait for it to come back up, I've never had a case where I had to do a manual 
repair.

Larry Troxler