FC Tape Profile Position
Stewart Wyatt
stewart at hpbs3928.boi.hp.com
Fri Nov 14 07:16:24 PST 1997
* From the T10 (formerly SCSI) Reflector (t10 at symbios.com), posted by:
* Stewart Wyatt <stewart at hpbs3928.boi.hp.com>
*
Tape Profile Position of HP CPB
Stewart Wyatt, Controller Hardware, Hewlett Packard CPB,
Boise, Idaho.
(CPB is Hewlett Packard's tape drive division.)
November 13, 1997
At the tape profile meeting in Palm Springs, Dal Allan
suggested that the participants post their positions and
assumptions relative to the selection of class 2 or 3. I think that
one of the problems that the working group has is that we have
not had a consistent set of assumptions and expectations. I have
been in favor of providing a profile for both classes of service
though I intend to implement class 3 first. I would like to
support that decision by answering some of the assumptions that
I have heard promoted in the profile discussions and presenting
some of my own assumptions.
1. Fibre Channel has to reach cost parity with SCSI for it to
become a widely accepted interface.
I hear this assumption everywhere Fibre Channel is discussed
except in the tape profile meetings. When Seagate introduced
the first Fibre Channel drives, they advertised that they were at
cost parity with Differential SCSI and would be at cost parity
with Single Ended SCSI by the second year. Our marketing
data is that most customers will not switch from SCSI to Fibre
Channel without the products being at cost parity and a
compelling performance advantage. It will, of course, require a
large volume of products to drive the price down. This creates a
"catch 22" situation. Adding expensive features to all products
not required by the large volume applications will increase the
price of the products and slow adoption of the standard.
2. Creating a class 2 profile for tapes will influence disks to adopt
class 2.
This argument is without any historical precedence. Last year
(1996) 106.7 million disks shipped generating $25.2 billion in
factory revenue while 5.2 million tapes shipped generating $3.0
billion in factory revenue. The tape industry is so small that it
has little influence with component vendors. The fact that we
are starting a tape profile several years after the disk profile
started demonstrates how tapes follow disks. With its smaller
volumes and revenues, the tape industry has relied heavily on
leveraging disk industry developments to make new product
development cost effective. Most tape companies are not
vertically integrated and are not able to develop the hardware
(asics) for interface solutions. A typical tape product today is
using SCSI hardware developed originally for a disk drive
application.
There are interface solutions available that support multiple
classes of service that are designed for host bus adapter
solutions. Typically these parts have a PCI interface on the
backend to enable direct access to host memory. These products
cost too much, are too large and consume too much power for
consideration in a peripheral product. Typically tape
applications buy a "stand alone" interface chip or integrate an
"interface core" with other functionality into a tape controller
ASIC. All of the available designs, without exception, are
strictly designed around the PLDA requirements. None of the
suppliers have any plans to support class 2 operation in
hardware. The relatively small volumes of the tape industry
make it very difficult to interest an ASIC supplier to develop
hardware that has unique characteristics for tapes.
3. Class 3 offers the fastest path to market.
As stated in the previous paragraph, class 3 designs are available
and proven while class 2 hardware has to be developed giving
class 3 an enormous time to market advantage. Class 3
solutions exist which have been proven in disk applications
dramatically decreasing risk and development costs.
4. Class 2 offers better support for queuing and fabric support.
This position is widely held by people who ought to know. The
issues are out of my range of experience. I assume that the class
2 promoters know what they are talking about. The question
that comes to mind is how soon these types of applications are
needed. I don't think that these requirements will occur in the
next few years. I think the current applications are more like
locally backing up a RAID box on a server.
5. Tape manufacturers all support class 2.
While some manufacturers support class 2, HP, among others
supports class 3. In the Palm Springs meetings several manufacturers
voiced support for a class 3 profile.
6. Adopting class 3 precludes class 2, or its corollary, we only
want to support one class of service.
The first statement is only true if class 3 meets all of the market
needs and no large customer or group of customers requires
class 2 service as a purchase requirement for their application.
The corollary is purely wishful thinking. Those who stay in
business will support whatever features their customers require.
Customer requirements will change over time. We supported
asynchronous SCSI until Fast SCSI came along. After
developing fast we implemented wide. Every generation of
products incorporates new controller features as soon as there is
enough market demand to make them cost effective. In a few
years products may support both classes of service to allow
systems integrators a choice - we offer SCSI products that
support a large number of interface options today, why will
Fibre Channel be different?
Conclusion:
I have been involved in the tape profile discussions since they
began. The arguments for selection of class of service have all
been based on technical issues. I hope that this memo brings
some implementation and market perspectives to the
discussions.
The proponents of the class 2 and class 3 approach are deeply
entrenched in their respective positions. Additional discussions
are unlikely to cause anyone to change their mind. There are
significant merits to both approaches which has brought the
effort to choose one class of service exclusively to a complete
deadlock. Class 3 solutions are already shipping and will
continue to be shipped regardless of any decision made by the
standards community.
Realistically the only decision left to the standards community is
whether profiles supporting both classes of service will be
developed by the standards group or will be developed
privately. I strongly favor developing profiles supporting both
classes of service in the working group. Adequate time has
already been spent discussing the relative merits of the two
proposals. Delaying product implementation for continued
debate will only harm the industry and delay or imperil to
acceptance of Fibre Channel.
--
Stewart Wyatt stewart_wyatt at hp.com
CPB Boise Controller Group Phone: (208) 396-3594
Hewlett Packard CPB Boise, PO Box 15, MS 477, Boise, ID 83707
*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at symbios.com
More information about the T10
mailing list