Subject: | |
From: | |
Reply To: | |
Date: | Fri, 31 Mar 2000 12:15:11 -0800 |
Content-Type: | text/plain |
Parts/Attachments: |
|
|
[public response to private message that brought up an interesting
topic]
> The question is, how well are the next generation of tape
> drives going to work.
> I keep thinking the I/O on HP3000s is pretty slow, at least
> on paper. I know MPE
> is so efficient that it doesn't generate as much I/O to
> accomplish the same tasks,
> but with PCs getting 160mb a second ultra scsi, you would
> think I could get
> something faster than 20mb a second on my old HP3000s.
Yes, HP's current I/O has been overtaken by the times, which is a
reason (probably among others) why they're going to PCI for future
hardware.
But don't be too impressed by numbers like 160MBytes/sec for Ultra160
or 400MBytes/sec for dual-loop 2Gb FC, because no drive in existence
can deliver even a quarter of that. The SCSI speed specifications are
for the *bus*, not for any particular device. The fastest drives
currently available, like Seagate's 15,000RPM "Cheetah X15" models,
have a maximum off-the-platter transfer rate of 47.4 MBytes/sec. And
that is only on the outer tracks. And can be sustained for a maximum
of one rotation: 4 milliseconds. Add in directory reads and
file-fragment seeks, and it's easy to see why most systems can't keep
a 5MByte/sec drive fed. Wait until we get to multi-petabyte
holographic storage on a SlamminSuperUltra5000 interface; *then* the
tape manufacturers will be the slow link in the chain. Though we'll
probably be using something else by then anyway. Until that time, it
will take something like a 4-spindle non-parity stripe set of
high-speed drives on a high-speed bus to make a DLT8000 break a sweat.
Steve Dirickson WestWin Consulting
[log in to unmask] (360) 598-6111
|
|
|