Subject: Re: AHA-2940 UW SCSI adapter problems?
To: Andy Ruhl <acruhl@gmail.com>
From: Chuck Swiger <cswiger@mac.com>
List: netbsd-users
Date: 07/19/2005 12:57:21
Andy Ruhl wrote:
> On 7/19/05, Michael Parson <mparson@bl.org> wrote:
>> With your 2940UW, you're not going to see as many benefits over today's
>> ATA stuff.  IIRC, UW was 40m/s, so even an older ATA100 disk would boast
>> higher burst throughput on a single device.  Loaded with 7 disks though,
>> the SCSI would show what it was made of. =)

A single SCSI disk at UW speeds compares quite well to all but the most recent 
ATA100 or ATA133 drives, in part because the SCSI drives tended to have a lot 
more cache than normal ATA drives.  SCSI is similar to the "SE" ATA models like 
the WD1200SE.

They also tend to spin at a higher RPM: 10K or even 15K for SCSI, versus 7200 
or maybe 5400RPMs.  Saving 3ms due to rotational latency makes a huge 
difference, especially to non-sequential or multithreaded access patterns.

> Not only that, but aren't the claims of throughput rates on ATA
> devices between the on disk cache and the controller or something? No
> way do these devices actually move 133 megs a second sustained
> (talking about ATA133), at least from what I've seen on a machine that
> should have been able to handle it.

The ATA numbers reflect the bus speed, not the maximum possible transmission 
rate of data.  In practice, you typically can get up to around 90% of the bus 
throughput as data; ie, you can only get around 90 Mbs of useful data through a 
100Mbs ethernet link once things like the protocol and ethernet frameing 
overhead are taken into account.

A good ATA133 drive will deliver 45-50 Mbs sustained sequential, which is 
marginally better than the max one could get through a UW-SCSI card.  Modern 
ATAPI devices are also using (including? inheriting?) a large portion of the 
SCSI command set, and things like SATA NCQ are derived from SCSI's TCQ, even 
though they are using a completely different physical/eletrical connection, so 
the differences are converging to an extent.

A 5-year old SCSI drive strikes me like a Mercedes or a Caddy which has just 
rolled over 100K miles: it's used, broken in enough to be comfortable, but it's 
still got twice as much life left in it, give or take.  A 5-year old ATA drive 
is not common, and the few I have strike me as less than convincingly reliable. 
  (They tend to go as scratch space for a web proxy server, or for a testing 
box, or some such.)

-- 
-Chuck