[ale] ext3

[Stuffed Crust]

On Tue, Sep 18, 2001 at 10:51:12PM -0600, Chris Ricker wrote:
> Actually, that's the common perception, but it also happens to be wrong.  
> If you benchmark ext2 and ext3, for example, you'll find that 3 gives higher
> throughput than 2 under almost all real or contrived loads.

READ speeds will be about the same, all else being equal.  However, 
WRITE speeds are the ones affected.

It depends on what level of journaling you're using -- If you journal
file data as well as metadata, I can guarantee you're going to have much
slower write speeds.

> The reason is that, while it's true that journals end up writing metadata
> twice (almost no jfs actually journals the files themselves), they also end
> up coalescing the writes, minimizing head motion on the drives....

It depends greatly on the loads, doesn't it?  Under light loads, there's
no perceptible difference.  Under heavy loads.. ther's going to be
slowdowns.  And obviously, the more cache/buffer space, the better.

I have point you to a bank of machines I'm keeping tabs on -- they have
a 5x73G hardware RAID5, and no write caching at all.  writes are much
slower than ext2... but a 15 second recovery (vs several hours for ext2)
is worth the tradeoff, given that we read far more often tan we write.
(Oh, these machines tend to be fairly loaded)

 - Pizza
-- 
Solomon Peachy                                    pizzaATfucktheusers.org
I ain't broke, but I'm badly bent.                           ICQ# 1318344
Patience comes to those who wait.
    ...It's not "Beanbag Love", it's a "Transanimate Relationship"...
--
To unsubscribe: mail majordomo at ale.org with "unsubscribe ale" in message body.