Performance 4.x vs. 6.x

[Mike Horwath]

On Sun, Oct 15, 2006 at 07:03:04AM -0700, Danial Thom wrote:
> We come from Earth; we're just more informed. The
> WD740ADFD's do NOT work on Freebsd 4.x. I'm sure
> you are talking about the WD740GD. I DID say the
> NEW ones. They are a lot faster than the GDs. I
> used them with FreeBSD 4.x with an Areca SATA
> card and they performed similarly to 15K Fujitsu
> SCSIs on a heavily loaded squid system. 

So you required a SATAII controller for an SATA disk?

squid?  Sure, somewhat heavy access.

How about UseNet?

I have 16 WD740ADFD disks in a RAID5 config (with a hot spare even!)
that is just cranking along.  Economics said I had to go this route,
not performance.

Since I needed *more* spindles and needed *more* space, I had to go
this route.  It performs adequately.

It replaced the 6x73GB disks at 15KRPM - performance was *much* better
with the SCSI, but at 2.7TB per day, I needed *more* retention to feed
back to my peers and my internal network.

> Also when you are using a SATA drive that *works*
> in 4.x its running in some reduced transfer mode,
> so you can't expect to get optimal performance or
> anything close. So "works" is almost a euphemism
> for "doesn't barf", but they don't really work
> well no matter how good the drive is.

More spew.

> For my needs, its cheaper to go with SCSI than to
> buy the sata card as there's only about a $250.
> difference in the SCSI  hardware, and I don't use
> up my slot. Hence, the silliness of operating in
> the FreeBSD camp.

Ah, cheaper, now we know the reason for the difference.

It sure isn't performance.

> Note that the drives work with 6.x but squid
> performance doesn't measure up, so again, 4.x
> with SCSI is the best bang for the buck choice.

Change controllers and don't be a wank.

I went down the path of ARECA controllers for my news spool project
(30TB of disk) and the controllers were fast for access, but randomly
dropped disks or volumes for no reason I could find.  I loved the
speed, hated the randomness.

I changed down the AMCC path, I lose some performance (not worth
posting numbers, it was under 5%), but no more issues.

FreeBSD 6, Opteron systems and 64bit.

The reason for this part of the post?  To show it isn't an issue of
not wanting to embrace new hardare/architectures, but that 32bit
operating systems and hardware isn't dead, for any of the reasons
given so far in this thread.

-- 
Mike Horwath, reachable via drechsau at Geeks.ORG