SSSWG - June 1994 Minutes
IEEE SSSWG Attendance June 1994
NAME COMPANY
_______________________________________ _______________________________________
Bob Baird Hewlett-Packard
Peter R. Berard Battelle, Pacific Northwest Labs.
LeRoy Botten Computer Sciences Corporation
John L. Cole US Army Research Laboratory
Bob Coyne IBM, Federal
Richard Garrison Martin Marietta Corporation
Dr. Bruce K. Haddon Storage Technology Corporation
Andrew Hanushevsky Cornell University
D. Creig Humes NASA Langley Research Center
Wayne Hurlbert Lawrence Berkeley Lab
Tom Jefferson Sandia National Laboratories
Phillip L. Keener Storage Technology Corporation
Steve Louis National Energy Res. Supercomputer Ctr.
Dennis Luck Department of Defense
Gary Mueller StorageTek
Jon W. Peake IBM Storage Systems Division
Michael T. Peterson Digital Equipment Corp
Ed Poole IBM Corporation
Rich Ruef IBM/FSC
David E. Skinner Fujitsu/ASG
Jodie Smith Storage Technology Corp.
Linda L. Tefend E-Systems, Inc.
Richard Wrenn Digital Equipment Corporation
The June 1994 meeting for the IEEE SSSWG was held at
Cavanaugh's Inn in Kennewick, Washington June 7-10.
Peter Berard (PNL) was the host company.
The meetings began on Tuesday morning with a group
session, Bob Coyne presiding.
The Wednesday evening dinner was planned for a winery
on the Washington-Oregon border. We needed to be at
Cavanaugh's at 5:30 on Wednesday to catch the bus for
the trip to dinner.
Bob Coyne said there was a Sam Coleman rumor that Open
Vision was closing its San Diego office and was dropping
Unitree as a product. David Martin responded that Open
Vision was consolidating the San Diego and Minnepolis
offices. Open Vision was not dropping the Unitree product.
And Open Vision was going to target lower-end business
opportunities (about 98% of their business) instead of
high-end businesses. Open Vision has bought DISCOS and
is adding NSL features to Unitree.
The August 1994 IEEE SSSWG meeting will be held
August 9-12 and will be hosted by Fujitsu (Dave Skinner)
in Longmont, Colorado.
The October 1994 SSSWG meeting is tentatively scheduled for
October 11-14 in New Orleans. Diane Garrison (Rich's wife) works
for a travel agency and will look into reservations. The October
meeting may be rescheduled if necessary to not coincide with
the October POSIX meeting.
ACTION ==> There was an action to determine the dates of the October
POSIX meeting.
The December 1994 SSSWG meeting is tentatively scheduled for
December 6-9 (or 13-16??). Possible locations for this meeting
are Boca Raton, Florida, San Diego, California, and the
bay area (Cupertino Convention Center - Bob Baird).
The current version of the Reference Model is now frozen.
Coyne will get with Andy Hanushevshy and Rich Garrison this
week to clean-up the document so it can be taken to ballot.
A copy of the Reference Model, along with a ballot sheet,
will be sent to each voting member of the SSSWG. The ballot
will indicate that the recipient has 30 days (+5 mail days)
to respond. A vote was taken as to whether no response would
be interpreted as a YES vote or as an abstention. It was
BOB==> decided that no response would be interpreted as a YES vote.
Creig Humes said there were 36 voting members.
There was additional discussion about ways to motivate
voting members to read the reference model. Suggestions were
made to give away t-shirts or to include a question on the
ballot that asked if the voter had read the document from
cover to cover.
It was brought up that any substantive changes to the reference
model would be held for the next version. Only inconsistencies
and grammatical errors would be corrected in the current version.
Creig Humes said there were 36 voting members.
POSIX standards were the next topic for discussion. Bob Baird
said the POSIX activity was too ad hoc and that 1244 should not
support POSIX implicitly.
Coyne countered that POSIX wants to mount tapes and that some 1244
folks want to help POSIX mount tapes. The idea of pulling the
groups together is to mitigate the impact of the two groups
working against each other. Currently, SSSWG 1244 members from
STK, Dec, and Fujitsu are active in both groups. This puts some
storage experts in the POSIX group. Having storage experts in the
POSIX group should limit the ability of a single company
(Fermilab) from driving the POSIX standard to baseline the
Fermilab product.
Both POSIX and IEEE SSSWG are now aware the other group exists.
If SSSWG thinks the POSIX work is overlapping/redundant to the
1244 effort, we can challenge the POSIX standards work.
If an individual in SSSWG is concerned about the POSIX standards
activity, that individual should try to support both groups.
Mike Peterson indicated that he had given the POSIX group the
PVL requirements and was waiting for a response.
Individual subcommittee chairs then presented their goals
for this week.
PVR - Dave Skinner
The Tahoe meeting was not especially productive.
They intend to resume work this week and forge ahead
with the DEC model. The PVR group needs nothing
additional from Andy H and assume that environmental
are being handled.
The PVR group will attempt to reestablish a schedule
this week and expect to propose a ballot date for the
PVR standards document by the end of the week. Merritt
Jones will put a ballot team together and needs 4 to 6
month lead time to get the ballet team in place.
The PVR group proposed that a draft ballot be taken first.
Andy H. agreed that the environtal issues are handled.
Mover - Creig Humes
No one is able to work on the standards effort between
meetings so progress is slow.
Coyne indicated that there was a push by DOD, Cornell
to get mover standards out and wants a ballot date
by October.
PVL -Mike Peterson
Expects to have a draft standard by Wednesday. The PVL
interfaces with the PVR, the Mover, and the Storage
Server.
Mike expects to continue to progress and hopes to catch
up with the PVR and Mover subcommittees.
There was a discussion about how long between a draft document is
ready and a draft ballot is available.
BOB==> I didn't get the resolution to this question if there was one.
Storage Server - Bob Baird
Greg Nuss has changed companies and is unable to
continue to support IEEE SSSWG.
Larry Pelletier has lost his travel funding.
Bob Baird is the only Storage Server member present
this week. He intends to talk with other subcommittees
about storage server interfaces with the other
components of the model.
SSM - Steve Louis
Leroy (last name ?) is replacing Carolyn Buford.
The SSM group has rewritten the SSM section for the
Reference Model using Jodie Smith's idea. Thomas
Jefferson has read the rewrite and does not disagree.
The outline of the SSM standard has been ready since
the December 1993 meeting in Las Vegas.
Coyne asked how the interface between the SSM and the
PVR is going to be expressed. Wrenn replied that the
SSM would access the PVR the same as any other client.
Humes (MVR) added that he has always assumed that no
special interface was required to interact with the SSM.
The question was asked if there would be enough
information to support security requirements, that DOTS
would use a client API with authentication/authorization
control for the SSM.
Rich Garrison then mentioned that OMG was requesting RFT
responses and suggested that 1244 put a cover letter on the
Reference Model and submit.
Break-out room assignments were then made.
An announcement was made that lunch would be served in the
Atrium room promptly at 12:00.
The next group grope is scheduled for Wednesday at 3:30.
Bob Coyne is leaving Thursday morning.
Ed Poole is coming in Tuesday afternoon.
The group then broke into subcommittee meetings.
Thursday, June 9, 1994
A group meeting was convened at 1:00 with Andy Hanushevsky
presenting.
-------------------------
Slide 1 - General Environmental Service Issues
- Communications
+ RPC Support
+ Message Passing Support
+ Broadcast and Multicast Support
+ Asychronous Event Support
- Transactions
+ Distributed Database Support
- Location and Naming
+ Object Location Issues
- Security
+ Authentication
+ Authorization
+ Enforcement
+ Auditing
+ Data Encryption
- Management
+ Alarms, Monitoring, and Logging
+ Configuration Management
+ Policy Management
+ Internationalization
A profile of the environment is:
- things outside the purview of the dots.
- things that can be purchased
- things that work well together.
Maybe internationalization needs to be its own separate bullet
item (Wrenn).
If the PVR passes an error code to the PVL, how does the PVL
translate the error code to human readable information. Should
there be a common method of translating error codes to text.
(Wrenn) Hanushevsky agreed that this is an environmental issue.
----------------------------------
Slide 2 - Communications
+------------+ +-------------+
| |<------Control Path-------->| |
| Server 1 | | Server 2 |
| |<-------Data Path---------->| |
+------------+ +-------------+
- Communication Needs (OSI, SNA, TCP/IP)
+ Asychronous events (send and notify)
+ Non-blocking communications
+ Automatic Protocol Version selection
+ Authenticated Communication Paths
+ Streamed Data
- Communication Possibilities
+ Authenticated RPC for control path (DCE or ONC+)
+ Proprietary data path allowed
+ Still need solutions for:
= Asychronous events
= Non-blocking control communications
Streamed data // multiple calls // data "put together" //
client sees a single stream of data.
Assumes control path does not carry large amounts of data.
The PVR status may include large number of cartridge IDs.
May need streamed data of client receives status in chunks.
The other solution is to rely on the Data Path.
-----------------------------------
Slide 3 - Transactions
+---+-----+ +-----+
| C | | | |
| L | MVR |<--------------------------->| MVR |
| I | | | |
| E |-----+ +-----+
| N | +-----+
| T | | |
| |------------------------------------------->| PVR |
| | +-----+ | |
| | | | +-----+
| |---------------------------------->| PVL | ^
| | +-----+ | | |
| | | | +-----+ |
| |----------------------->| VSS | ^ |
| | +-----+ | | | |
| | | | +-----+ | |
| |---------->| NS | ^ | |
| | | | | | |
+---+ +-----+ | | |
^ | | |
| | | |
| | | |
v v v v
Database Database Database Database
- Data Consistency
+ Single database requires common standard database
+ Multiple databases need to be consistent
+ Need environmental distributed transation support
+ Can use X/Open XA standard for Transactional RPCs
+ Even with support, vendor needs to do a lot of work.
Intent is to encourage inter-operability. May have multiple
profiles, one specifying DCE RPC, another specifying ONC+ RPC.
(Wrenn, Skinner) think standards should pick a single technology.
Needs a single profile that is very specific about technologies.
May have multiple profiles. Each profile can be very specific.
-------------------------------------
Slide 4 - Location and Naming
+----------------------+
| |
+-------------------+-------------------+ |
| | | |
| | v v
External Name -----> Object ID ------------> Object Location
(Typically |
Hierarchical) |
| |
| Model Method
|
Typical Method
- Location and Naming (DNS, X.500)
+ Typically together (e.g., DNS, X.500)
+ Scalability drives naming structure
+ Wide area solutions have large granularity
+ Small object location under control of object servers
+ Emerging standards: CORBA, DOE, DSOM, OLE, OpenDoc, SOM
+ Very difficult to have location independent object identifiers
+ One approach:
+---------------------------+----------------------------+
|128-bit Obj Server Location|128-bit Obj Server Dfined ID|
+---------------------------+----------------------------+
+ OID subgroup will investigate.
(Wrenn) Suggestion to solve wide area problem now. Don't try to
solve small object location until later.
Cannot base anything in a standard on a non-public standard.
A profile is a recommended practice.
-----------------------------
Slide 5 - Security
+--------------------------+
+---+ | ENFORCEMENT |
| | Client-Object- | ^
| Grantor |-----------------+ | Capabilities or privileges
| and | AUTHORIZATION | |
| Client | Client-Object |
| Identities | Relative | ^
| |-----------------+ | Credentials
| | AUTHENTICATION | |
+--+ | Client Relative |
+---------------------------+
- Authentication
+ No one size filts all solution:
= Private key (e.g., Kerberos)
= Public key (e.g., RSA)
= RSA
= SmatCards, etc.
+ Credentials are reasonably standard but ...
= Accounts, Roles, Delegation, Impersonation, etc.
+ GSSAPI can be used to hide many variations
+ Claim that authentication is a communication level function
- Authorization
+ No one size fits all solution:
= Discretionary Access Control (POSIX ACLs)
= Mandatory Access Control (POSIX labels)
+ No common standard interface found
- Enforcement
+ Fairly straightforward
+ POSIX operation (probably sufficient)
+ No common standard interface found
- Audit
+ POSIX audit interfaces
+ No common standard interface found
- Data Privacy or Encryption
+ Haven't addressed the issue
-------------------------------
Slide 6 - Management
- Storage System Management Subgroup
+ Two main approaches
= CMIP
= SNMP
+ Need to also address
= Object definitions
= Policy management
= Configuration
= Internationalization
* PG PERF Monitoring *
+ Object management interface unsettled:
= Two interface approach (service and management)
= one interface approach (servicd + authorization = management
Internationalization and Two interface approach were removed from
slide 6 during the ensuing discussion.
-------------------------------
Slide 7 -
V M P P
L N S V V V
S S S R L R
(Standard Storage System)
+-----------------------------------------------------+
| Distributed Transaction Processing Layer |
+-----------------------------------------------------+
| Management Layer |
+-----------------------------------------------------+
| Security Layer (Authentication) |
+-----------------------------------------------------+
| Communications Layer |
+-----------------------------------------------------+
| | |
| | |
Client Client ... ... Client
-------------------------------------
Discussion (Louis, Hanushevsky) Standards documents can reference
a profile document. There is not a DOT for the profile document.
It is a profile, not a standard. A profile document describes the
environment. What level of detail is required in the standards do
not have to rewrite information included in the profile document.
Do not want the user/vendor to need both documents.
Discussion about not identifying anything in a server standard that
the environment cannot handle (asynchronoug events, for example)
(Wrenn). Can define context of asynchronous events without knowing the
method of putting the events on a wire. The DCE environment
supports this (Louis).
A profile document is background information for the
standards writers.
Discussion (Wrenn, Skinner, Ruef, Hanushevsky) The Dispatch Issue
Every layer is going to have to resolve this issue. Does that
make the dispatch issue an environment issue. Suppose you
want to get a handle, and then send a command. With RPC, can
have stubs with each stub representing a separate PVR. Each RPC
stub has the same entry points. Can use DNS. The dispatcher
must forward the command to the correct stub. The solution is
different for each RPC. Can use static binding or dynamic
binding. For example,
pvr_get_handle("PVR andy_PVR")
status = pvr_show(handle, &buffer)
The routing information is put into the handle when the get_handle
call is issued. Ruef==>DCE, ONC RPC puts the routing information
into the handle. This is a non-problem.
THE END
Linda Tefend
SSSWG Home