Re: Disk issue: More arms versus improved hardware

["Charly Jones" <charly301@xxxxxxxxxxx>]

History proves Larry's point.  In 1978 the 62PC "piccolo" drive
that was on the then just announced System/38 had an average service
time of 35 milliseconds, meaning you could get almost 30 unique chunks of 
data from one disk in 1 second.  The 3370 that came out a few years later 
had an average service time of 30 milliseconds.

Disk technology has come a long way in the last 33 years and now on the 
latest and greatest 18 gig 10,000 rpm drives like the 4318, 6718, 6818, 
8618, and the 8818 the technical specs I have show a read seek time of 4.9 
milliseconds and a latency of 3 milliseconds for a total of almost 8 
milliseconds.  The write seek time and the 9 gig drives are each a little 
slower.  So today, best case, you can get about 125 unique chunks of data 
from one disk arm in one second.  WOW -- four times as much in 33 years!!!

If you plot the disk service times of the latest and greatest disks over the 
years they very nicely fill in a graph with these two end points.  You can 
project it out into the future yourself...

So if you have code that does a SETLL and read or something similar, you 
need more than a few disk arms if you want good performance for a bunch of 
users.  The IOP can cache the writes, main memory can cache things that 
others have looked at recently and you are left with stuff that is on disk 
and needs to come into main memory in a hurry.  Disk technology isn't 
keeping up with processor technology and everything else.

Charly Jones
Geezer in Gig Harbor

>From: Larry Bolhuis <lbolhuis@arbsol.com>
>Reply-To: MIDRANGE-L@midrange.com
>To: MIDRANGE-L@midrange.com
>Subject: Re: Disk issue:  More arms versus improved hardware
>Date: Tue, 27 Feb 2001 10:27:47 -0500
>
>Rob,
>
>   Certainly you are correct. In theory if one drive could be made fast
>enough to handle the IO for a system then that's all it takes. The 10K
>8G units combined with the IOPs and IOAs of today's 270/8xx series will
>perform roughly 70% better than 4G drives with older controller
>technology. In other words 3 of these would perform about the same as 5
>4G units. (Move the same amount of data at the same utilization rate).
>
>   Without doubt the controllers will continue to get faster, bigger
>caches, and smarter code. Drives too will continue to get faster, but
>there is a limit with all things physical.  RPM will likely still climb
>some because platters continue to get smaller, this of course reduces
>latency.  Seek times are in a catch-22. If you want to seek faster you
>must push the heads harder to get them there quicker, but you must then
>strengthen the arms holding the heads to sustain the additional push,
>which usually makes them heavier, which means you have to push them
>harder....  Some guy named Newton comes to mind here.  Improved
>materials can help, but before we see any big improvements here I'd
>expect some new technology will need to be released.
>
>   - Larry
>
> > I've heard this argument that more arms is better quite often.  But
> > sometimes I wonder if it is true versus improved performance of the 
>newer
> > hardware.  For example, let's say you could get IBM PC original 10MB 
>hard
> > drives into a single level storage machine like the AS/400.  How many of
> > them would it take to give you the performance of the 8gb drive?  Or 
>could
> > it ever be done?
> >
> > I will admit the 17gb drive is better left for non performance critical
> > applications.  We use them on a secondary ASP for backing up our PC's 
>using
> > ADSM.
> >
> > Rob Berendt
>
>--
>Larry Bolhuis          | Cogito Ergo Vendo iSeries
>Arbor Solutions, Inc.  |
>(616) 451-2500         |               (I think, therefore I buy iSeries.)
>(616) 451-2571 -fax    |
>lbolhuis@arbsol.com    |        #3  1951-2001
>+---
>| This is the Midrange System Mailing List!
>| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
>| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
>| To unsubscribe from this list send email to 
>MIDRANGE-L-UNSUB@midrange.com.
>| Questions should be directed to the list owner/operator: 
>david@midrange.com
>+---

_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com

+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---