SNIA-OBSD: RE: Object oriented issues

Chris.Malakapalli at seagate.com Chris.Malakapalli at seagate.com
Wed Aug 16 13:10:16 PDT 2000


* From the T10 Reflector (t10 at t10.org), posted by:
* Chris.Malakapalli at seagate.com
*
Jim,

As part of the engineering group at Seagate research, we have explored into
many architectural issues relating to object storage. We intend to publish
some of the experiences we have gained as a result of getting our feet wet
in this space.  Hopefully, it can help those who have an interest to pursue
it further.

Regarding your comments on TCP/IP and using SCSI, I tend to agree with you
that object storage it is more closer to a simplified NFS rather than a
glorified disk drive.  Infact, the original CMU NASD model was developed to
run over TCP or UDP/IP. The initial version used DCE RPC and the client
looked more like modifed NFS.

Given the recent advancements in mainstream network bandwidths and
latencies, storage protocols running over storage specific interconnects
may no longer enjoy the edge they once did. Clearly, the industry support
for mapping block storage protocols over TCP/IP, VI and Infiniband are a
move in that direction. I think that SCSI makes sense as long as we are
running block oriented operations across the network.  Obviously, one would
want to leverage the existing block infrastructure, meaning the server file
system support.  But when it comes to object storage, the server file
system needs to change and as long as it needs to change, you have gained
yourself the architectural freedom to define a new protocol that does not
have to be SCSI.  However, for backward compatibility, you might want to
make sure that object protocols can run over SCSI, but I am not sure how
compelling the value proposition in this case is.  The real motivation
might be to design it well to run over TCP/IP, VI or IB so future
applications can take advantage of it.

-Chris






hafner at almaden.ibm.com on 08/16/2000 11:44:08 AM

Please respond to hafner at almaden.ibm.com

To:   dave.b.anderson at seagate.com
cc:   snia-obsd at snia.org, t10 at t10.org

Subject:  SNIA-OBSD: RE: Object oriented issues


Dave,

I started to write a very long winded reply to some of your comments, but I
think they are getting repetitive.

However, let me try to summarize what I think are our philosophical
differences and then summarize my position.

1) I look at an OSD as a simplified NFS, not a glorified disk drive.  You
look at it the other way around.
2) As a simplified NFS, TCP/IP is the natural protocol.  As a glorified
disk, SCSI is the natural one.

I don't deny that a SCSI specification and implementation is possible.  I
just think that it would have enormous barriers to timely standardization
and (more importantly) to pratical deployment.
TCP/IP has no fundamental architectural limitations; SCSI has many.
SCSI *might* be changed, but that will take time (both in the standards
process and in deployment of the infrastructures required to implement the
standard).
SCSI has a lot of excess baggage that isn't needed for a clean OSD model
(mode pages, to name just one).

Additionally, I don't see any point in SCSI's favor over TCP/IP in either
the short or long term .

You wrote:
 > I hope that I am willing as you to be convinced otherwise.
In fact, I don't really need to be convinced at all.  There are many things
that this industry is doing that I personally am agnostic about.  If some
things are going to happen anyway and I have opinions on "implementation",
then I'll speak up.  In other words, I'll stick my nose into anything with
an opinion, but I won't try to stop anything either.  (So, if you, et al,
want to push this through T10, I'll be willing to help, even if I think
you'd be better suited to push this elsewhere.  You'll notice that I didn't
make a big deal about this point in my talk at the T10 meeting, just raised
the banner.)

 >I also think your list of open issues is worthwhile.  Would be willing to
 >present a summary of your paper and your position on the protocol at the
 >OBSD working group meeting?
I think I can be persuaded.  Do I need to do anything in particular to get
on the agenda?

 >On the question of whether an OBSD is a disc drive, I fell pretty strongly
 >that it is, and I will try to line up somebody to address this at the WG
 >meeting.  (It turns out that there is a lot outside of OBSD that relates
to
 >this particular subject.)
If you're talking about packaging an OSD in something that looks like a
disk drive, that's not really the point.  The point is more how it will get
used.  An OSD's real value over a block storage device (aka disk) is in
sharing among many hosts.  That implies a network attachment.  That means
(today), two primary choices, FC (and so SCSI over FC) or ethernet (and so
IP or TCP/IP).  The physical dimensions are less important to me than the
fact that on OSD is a networked device!  Now we need to decide what's the
best network to use!

[.. deleted the rest of the history...]

Jim Hafner







*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org




More information about the T10 mailing list