P1394 Mail Archive: RE: P1394> SBP-2 vs. DPP 0.71 Cheat Sheet (?)

RE: P1394> SBP-2 vs. DPP 0.71 Cheat Sheet (?)

Atsushi Nakamura (atsnaka@mxa.mesh.ne.jp)
Tue, 24 Feb 1998 00:13:51 +0900

Greg, and all

Though I think this is good effort,
my impression of the list is that it doesn't
look too different from the requirement list
we have been discussing for the past several months.

I think ANY (good) transport protocol would support
most of the items listed in the cheat sheet section one way
or another, thus look alike if you compare it this way.
SBP-2 and Thin are both good transports because they
both support the list of requirements !
I don't think one is better, or worse than the other.

I think it is HOW each protocol supports an item
that make the difference, not WHAT they support.

This relates to the thinking behind the development of each
protocol and the intended applications or solutions.
If each one(SBP-2, thin) has their benifits, and people see the
benifits, (which I think is the situation now )
we don't necessarily HAVE TO decide the ONLY one just now.

( After all, the thin transport is only a few kilo-bytes....
No big deal to add a few more KB to 30(?)KB of SBP-2 ? (^^;)
Just joking ! )

Ats,

-------------------------------------------------------
Ats Nakamura
BJ Printing Technology Development 22
Canon Inc.

-----Original Message-----
差出人 : Greg Shue <gregs@sdd.hp.com>
宛先 : gregs@sdd.hp.com <gregs@sdd.hp.com>
CC : p1394@pwg.org <p1394@pwg.org>; pp1394@cpdc.canon.co.jp
<pp1394@cpdc.canon.co.jp>
日時 : 1998年2月21日 9:49
件名 : P1394> SBP-2 vs. DPP 0.71 Cheat Sheet (?)

>
>Hi all,
>
>The more I look at DPP 0.71, the more similar it looks
>to SBP-2. So, I put together a concept "cheat sheet"
>to see how the two really compared. I thought you
>might be interrested, so here it is... What do you think?
>
>- Greg Shue
>
>===============================
>SBP-2 vs. DPP v0.71 Cheat Sheet
>===============================
>
>There are many similar concepts between SBP-2 and DPP v0.71 which
>may not be well understood. The main difference between the
>protocols revolves around the a/symmetry of the protocol stacks.
>This "cheat sheet" is intended on laying the groundwork for a
>direct comparison of the protocol stacks.
>
>
>SBP-2 Model of Use for Comparison
>---------------------------------
>In order to get a reasonable comparison of the protocols, then
>a Command Set and Model-of-use must be outlined for SBP-2. Many
>of the details can be left out of the outline as long as the
>essence of the profile is maintained.
>
>Model of Use
>------------
>
> - cross-logins are established between the two devices.
> (This avoids a merged target/initiator driver)
>
> - each device provides a fetch engine and a task list.
>
> - each task list contains commands to transfer data in one direction
> (we'll choose it such that only WRITE transactions occur for
> moving the ORB data block)
>
> - each ORB allows one "data unit" to be processed.
>
>Command Set Provides For
>------------------------
> - cross-login information
> - negotiated max data block size per task per direction.
> - indicating byte-size of completed packet
> - carrying of Command IDs in CDB (2 bytes)
> - carry "SDU segmentation number" in CDB (4 bytes)
> - carry "SDU end of segment flag" in CDB (1 bit)
> - carry routing info in CDB (2 byte Dest PID, 2 bytes Src PID)
>
>
>Cheat Sheet
>===========
>
> CSR structures Thin Protocol SBP-2
> -----------------------------------------------------------------------
> Access Control i/f Connection Reg Mgmt Agent Reg
>
> Response Space SDU Response Space Fetch Agent Regs
>
> Completion status SDU Control Space Status FIFO
>
> Transfer block SDU Register data block referenced by
ORB
>
>
> Functional layers Thin Protocol SBP-2
> -----------------------------------------------------------------------
> Connection mgmt Thin Session Management Agent
>
> Data Transfer eng Thin Transaction Fetch Agent
>
>
> Session Mgmt Thin Protocol SBP-2
> -----------------------------------------------------------------------
> Access Request S_Connect login
>
> Reconnect T_Reconnect reconnect
>
> Execute App Cmd S_Command ORB exists on task list
>
> Access Release S_Disconnect logout
>
> Error Notification <none> Mgmt-ORB specified Status
FIFO
> completion notification
>
>
> Data Transfer Thin Protocol SBP-2
> -----------------------------------------------------------------------
> Transfer Space SDUAck ORB exists on task list
> Available (write implies avail) (existance implies
available)
>
> Transfer completion SDUComp Notification write to
> notification Status FIFO
>
> Transfer Abort from SDUStop Abort Task (Set)
> data receiver
>
> Transfer Abort from SDUStop Notification write to
implicit
> data sender Status FIFO - Abort
>
> Error Notification T_error Notification write to
implicit
> Status FIFO - error ind.
>
>
>Feature Comparison
>==================
>
>Feature Thin Protocol SBP-2
>-----------------------------------------------------------------------
>Each node can send/rcv Yes Yes
> data
>
>Dynamically allocates Yes Yes
> CSR Reg space when
> estab. connect
>
>Selectable Transfer Yes Yes, requires Create
Stream
> Mode request and separate
> task list
>
>Symmetric Structure Yes Yes
>
>Negotiated Data Unit Yes Yes
> size independently
> per direction
>
>Isochronous Support Yes Yes
>
>Multiple Cmd Support Yes Yes
>
>Bidirectional Cmds Yes Yes
>
>Multiple Independent Unspecified (possible Yes, via Multiple LUNs &
> Connection Support w/ multiple UNIT Dirs) multiple UNITS
>
>Session segmentation Yes Yes, in CDB
> support
>
>
>Distinctive differences:
>
>Feature Thin Protocol SBP-2 w/ proposed profile
>-----------------------------------------------------------------------
>Reconnect Support 2 sec window; 1 sec window, proposal
made
> either end may for programmable
extension;
> establish both ends must reestablish
>
>SDU Available SDU Ack timeout exists None (?)
> timeout Does this imply a real-
> time processing req. on
> whatever data is sent
> in an SDU?
>
>Connection Process 2 block writes 6 block writes (probably)
> (2 logins @ 2 per login,
> 2 for negotiated
params)
>
>Shared memory buffer Fixed location during Relocatable, partitionable
> connection
>
>Data Transfer more efficient for More flexible and
standardized
> data < MTU size
>
> data < MTU size SDUAck + WR Agent Reset + WR ORB
ptr +
> SDUComp + ORB prefetch + data xfer +
> SDUAck Completion Notify
> (3 transactions vs. 5 transactions)
>
> MTU size << data < SDUSend + WR Agent Reset + WR ORB
ptr +
> SDU size SDUComp + ORB prefetch + page table
+
> data xfer +
> SDUAck Completion Notify
> (SBP-2 costs 3 extra transactions,
> but a small overhead
>
>--
>Greg Shue
>Hewlett-Packard Company
>Office Products Division gregs@sdd.hp.com
>----------------------------------------------------------------
>