Subject: Re: ccd performance
To: Brian Hechinger <firstname.lastname@example.org>
From: Hauke Fath <hauke@Espresso.Rhein-Neckar.DE>
Date: 06/14/1997 13:15:32
At 5:02 Uhr +0200 14.06.1997, Brian Hechinger wrote:
>Dave Huang drunkenly mumbled...
>> Do you think things would be better if I made the ccd across a 2 gig Atlas
>> and a 4 gig Atlas? Both 7200 rpm, same # of cylinders and sectors/track,
>> but the 2 gig has 10 heads and the 4 gig has 20.
>> Or is disk striping pretty much useless unless you have identical drives?
>you could use these two disk yes. to get optimal performance you would have
>to sacrifice half of the 4G disk. make a slice on the 4G that is the same
>exact size as the 2G with a single slice on it. also, have the two drives
>match as far as offsets go. this would get you the same performance as having
>two 2G disks.
Not quite. Remember, the time of 17 sec/tr ST506 drives is utterly and
forever gone. ;)
"Half of the 4GB disk" is about the outer third of the disk's cylinders.
With zone bit recorded disks, the physics of this area (spoken in terms of
data density/bits per inch and data rate/bits per second) is entirely
different from a full 2 GB disk. With a SCSI ZBR recorded disk you have *no
way* of accessing "half the heads" to mimic half the capacity as the
internal data layout is entirely up to the disk.
And that's a good thing, too, even if UN*X-like OSes (and part of their
users) still cling to a fifteen-year-old notion of disk technology.
I.e., if you want to distribute load symmetrically, you have to get disks
with identical *physical* disk layout and make sure you don't break this
identity with partitioning.
>"Color is for web wanking. mono is for writing code. Gimme a Mono Sun 3 and
>an expresso machine any day :-)"
> -Bob Beck
Greetings from a Macintosh greyscale full page display. =8)
"It's never straight up and down" (DEVO)