<NIS.NSF.NET> [IMR] IMR86-07.TXT
JULY 1986
INTERNET MONTHLY REPORTS
------------------------
The purpose of these reports is to communicate to the Internet Research
Group the accomplishments, milestones reached, or problems discovered by
the task forces and contractors in the ARPA Internet Research Program.
This report is for research use only, and is not for public
distribution.
Each task force and contractor is expected to submit a 1/2 page report
on the first business day of the month describing the previous month's
activities. These reports should be submitted via ARPANET mail to
Westine@B.ISI.EDU.
Reports are requested from BBN, LINKABIT, ISI, LL, MIT-LCS, NTA, SRI,
and UCL.
Other groups are invited to report newsworthy events or issues.
BBN LABORATORIES AND BBN COMMUNICATIONS CORPORATION
---------------------------------------------------
DIAMOND MULTIMEDIA COMMUNICATIONS SYSTEM
We have developed a new release (2.0) of the Diamond System. Included
in this release are fixes to bugs as well as several new features,
including:
* X.400 message transport system for multimedia messages: This
system is completely compatible with the MPM in that it obeys the
MPM protocol if a message is directed to a host that runs MPM
rather than X.400
1
* Printing of documents: Support for printing Diamond documents
on Imagen 8/300 laser printers.
* MMConf Multimedia Conferencing Prototype: A prototype
multimedia conferencing system that allows multiple participants
to confer over Diamond multimedia documents.
Requests for obtaining the Diamond system should be sent to Kathy
Huber (KHuber@A.BBN.COM).
VAX UNIX NETWORKING
In the month of July, work continued on the port of Stanford
University's internet multicast software to 4.3 BSD Unix. A new
protocol specification was received from Steve Deering at Stanford,
and the port has been updated to meet the specification. Debugging
and testing of the software is still pending the arrival of 4.3 BSD
Unix at BBN.
WIDEBAND NETWORK
On July 15 and 16, a new Butterfly gateway was installed at the ISI
Wideband Network site, connecting the ISI BSAT and a local Ethernet.
The BSAT was configured so that the new gateway could interface to
the BSAT via a 2 Mbps connection. This configuration was chosen to
support experimentation with NETBLT protocol based file transfers
between ISI and MIT via the Wideband Network.
During the same ISI site visit, the ISI Voice Funnel machine was
upgraded to exclusively use 1-Mb Butterfly processor nodes and a VLSI
Butterfly switch. The ISI BSAT was also upgraded to use 1-Mb
processor nodes. All BSATs other than those at CMU and M/A-COM
Linkabit are now 1-Mb node machines.
On July 25, the DCEC BSAT and the DARPA Wideband Butterfly Gateway
machines were shipped. Installation is scheduled for August 4. The
BSAT and gateway will be connected via a T1 terrestrial link.
The maximum datagram message size for the Wideband Network was
increased to 2048 bytes. The increase is supported by the network's
use of QPSK modulation and datagram fragmentation on the satellite
channel. The Butterfly Gateways will implement the new maximum
message size.
A facility for fixed group (multicast) addressing was implemented and
will be included in the next BSAT software release. This feature
will support 3-(or more) way conferences between Wideband sites.
2
GATEWAYS
We completed testing a new release of software (Rel. 3.2) which we
believe will fix the problems encountered in the SATNET Butterfly
Internet Gateways. We have been running this new software in two
operational gateways at BBN for about a week and are planning to
install it in machines at other sites this week. After it has run
sucessfully in the CSS (Arpanet-Satnet) gateway we install the new
software in the European Satnet gateways (NTA, RSRE, CNUCE, and UCL).
SATNET
This month, the SATNET continued to be stable. Joachim Kaiser
(Kaiser Inc.) began working with Linkabit and COMSAT to assure a
supply of spare PSP modems. There were some problems in the links to
the local user sites. In mid-July there was a brief break in service
between the SATNET and the ARPANET when both US gateways went down
due to air conditioning and power problems. Also, since the new
Butterfly gateways are going through their first months of real-live
SATNET use, connectivity between the SATNET and the local networks at
RSRE, CNUCE and NTARE has been variable. The PDP-11s at DCEC and UCL
have continued to provide reliable service.
Bob Hinden
ISI
---
Internet Concepts Project
Paul Mockapetris attended the Internet Engineering Task Force
Meeting at the University of Michigan, 22-24 July.
Annette DeSchon made a number of performance improvements to the
XEROX TCP and FTP implementations which run in the XEROX
Development Environment on the 8010 workstation. She is also
developing a domain name resolver which will be accessed via the
standard software interfaces in the XEROX DARPA protocols package,
as well as by a special purpose test tool running in the XDE.
This "NameTool" may be used to send queries to the various domain
nameservers in the Internet, and to display their responses.
Multimedia Conferencing Project
Brian Hung completed the first application on the IBM-PC AT
workstation based on the newly specified set of data and control
structures. This application allows a user to scan, clip and
display documents on the high resolution screen continuously. The
user specifies the contrast and brightness parameters for the
document to be scanned, the top left hand coordinates and the
3
width and length of the new frame to be clipped from the scanned
document and the top left hand screen coordinates of the top left
hand corner of the new frame. The next step is to develop more
functions (e.g. write_to_file, read_from_file) and parallel
control structure (i.e. to enable simultaneous outputs to two
procedures) to allow a more varied set of applications to be
generated.
Supercomputer and Workstation Communication Project
Alan Katz is currently attending the Super Computing Workshop at
Cornell University.
The installation of the Butterfly Gateway to the Wideband Net at
ISI has allowed us to resume testing of FTP transfers across that
net. Unfortunately we're still bumping into apparent 16K window
size limits in host TCP implementations; investigations continue.
Meanwhile we are also testing raw capacity with ICMP echo request
packets which have no window or timeout constraints. Further
NETBLT protocol testing is planned for next month as well.
Steve Casner
Computer Center
No internet-related progress to report.
LINCOLN LAB
-----------
No Internet-related progress to report.
Jim Forgie
LINKABIT
--------
1. The NSFNET Backbone is now alive at six of the planned seven
sites. Lots of folks at the various supercomputer sites,
especially Hans-Werner Braun at the University of Michigan, worked
feverishly to solve or at least contain the bizarre bugs that
still munch on several campus networks.
2. Hans-Werner and I managed to confirm that the core system responds
to EGP hop counts, chosing the path of least total hops, where
total hops includes the contribution from the EGP hop-count field
provided by the non-core gateway. Our experiment involved two
gateways between the NSFNET Backbone complex and the ARPANET. One
gateway specified a hop count of zero and the other one, with the
effect that the core used the zero-hop path. Then the zero-hop
4
path was disabled, with the effect the core used the one-hop path.
This provides for a normal and backup path between the Backbone
and ARPANET, but is considered only an interim solution.
3. I attended the Internet Engineering Task Force meeting and
discussed several issues arising from the NSFNET activities. I
also raised many of these issues in messages exchanged via the
TCP-IP list.
4. The CNUCE (Pisa, Italy) fuzzball sputtered briefly to life and
then fell victim of a broken disk. Blasco Bonito of CNUCE, Roki
Rokitanski of DFVLR and I managed to confirm reachability of the
Italian swamp via the Fucino SATNET station.
5. I began working on routing mechanisms for the new DARPA/RADC
Multiple-Satellite System (MSS), which is a wonderful can of
worms. I constructed and am studying a number of scenarios, some
based on traditional ARPANET-style routing, some considerably more
bizarre.
6. I rebuilt the fuzzball namesolver and nameserver to improve
performance in the face of flaky Internet connectivity observed
recently. Our ARPANET access line also began behaving badly, which
is now under repair.
7. My next report will be dispatched from the University of Delaware,
where I will become Professor of EE starting about 1 September. I
intend to continue punching Internet bags as usual, so you may not
even notice the change.
Dave Mills
MIT-LCS
-------
No report received.
NTA & NDRE
----------
No report received.
PROTEON
-------
A lot has happened in the past several months, but I've been too busy
to make up reports. On the less technical side, Release 7.1 of the
gateway is out; it includes the network interface reliability
features mentioned in the April report. The EGP code from the PDP11
was also brought up. It also includes support for 1822LH/DH
5
interfaces and synchronous point-point links, through a variety of
level adaptors. The synchronous links only run at low and medium
speeds, due to restrictions in the Zilog SCC USART (which are
difficult to work with; we can lend a hand to anyone having problems
with these chips). Work programming a faster chip (capable of running
at T1 speeds) is well along.
Also, a great deal of work has been done to support extensive testing
in gateway production; a test rig in manufacturing completely
simulates the destination operating environment of the gateway,
allowing all the interface hardware to be tested with the operational
software configured for the actual network addresses it will run at.
Several other projects are also completed, and will be included in
the next release. A remote console capability (using standard TCP
server TELNET) is one, and cross network dumping of crashes to allow
fault analysis is another. We have also investigated higher
performance Ethernet boards; we have noticed that the Interlan
Ethernet board we are using could not match the performance of the
ring interface (which was admittedly designed for high performance in
light of early experience at MIT with high speed interfaces). We have
implemented a driver for the Excelan smart card, which is commonly
viewed as a high performance interface, and while it does have a
maximum bit rate three times higher with large packets, we did not
record any improvement on the packets/second figures. We are
evaluating the other commercially available Ethernet interfaces in an
attempt to find one which has the high throughput we want; most seem
to have confused 'smart' with 'fast'. We'd like to hear from anyone
with good experiences with high speed Ethernet interfaces.
Also, we have been seeing lots of problems with the RIP IGP
(distributed with Berkeley UNIX) in some of our larger and more
complicated customer sites. We are currenlty investigating this and
hope to have some results to report in an RFC and to the INENG group.
Finally, I attended the INARC meeting at BBN, 8-9 May.
Noel Chiappa
SRI
---
1. The final demonstration of the reconstitution protocols was
conducted on June 27 at Omaha. The F9 demonstration involved a
network-mobile host (SUN 150 onboard a SAC aircraft) flying from
a PRNET based around Camp Dodge, Iowa, into a PRNET based around
Omaha, NE. The SUN in the aircraft was in communications via the
ARPANET to a VAX at SRI. Because of antenna orientation, the
aircraft was isolated from both PRNETs for a short period of
time. The demonstration successfully showed the ability of the
6
RP gateways to track the SUN host as it moved from network to
network along with the end-to-end reliability of TCP.
Jim Mathis
UCL
---
1. Further work has gone into improving the performance of the C
version of our TCP. Some major bugs were found and fixed; this
improved the performance dramatically compared to what it had
been.
2. The UCL Internet probe (which uses ICMP echoes) was ported from an
LSI-11 on a Cambridge Ring to a SUN on an Ethernet.
3. A reverse ARP client was written for hosts running our C version
of the MOS operating system. This talks to any (release 3.0) SUN
RARP server.
4. Much effort has gone into our Rugby Clock experiment, which is
intended to develop a Time Service supporting the Network Time
Protocol. After some initial success, various problems have slowed
recent progress. Along the way, we discovered that the clocks on
our Olivetti and IBM PCs drift apart by up to 8 seconds per minute
relative to each other.
5. Two papers were submitted to journals: "Managing Heterogeneity in
Computer Network Interconnection" by Robert Cole, submitted to
Software Practice and Experience; and "OSI Transport Protocol -
User Experience" by Robert Cole and Peter Lloyd, submitted to IEEE
Communications Magazine.
6. Michael Wallace and Jerry Mulvenna of NBS visited UCL and talked
with Peter Kirstein and Steve Kille. The main topics were
SMTP/X.400 gatewaying in the light of UCL work, and the DoD
contract held by NBS for work in this area.
Peter Lloyd
7
TASK FORCE REPORTS
------------------
APPLICATIONS
No report received.
END-TO-END SERVICES
MULTICASTING and HOST GROUPS
A new RFC has been released containing a further definition of the
proposed Internet multicasting mechanism (see RFC966).
Steve Deering, "Host Extensions for IP Multicasting", RFC988,
July 1986.
THE INTERNET RESEARCH COMMUNITY IS URGED TO READ THIS DOCUMENT AND
COMMENT ON IT, as we intend to push towards its adoption as an
official extension to the Internet architecture.
Steve Deering has completed his second-version implementation of a
multicast agent under the Stanford V-kernel. This code is
available for distribution, and can be brought up on a dedicated
Sun workstation or Microvax II to to provide multicasting on a 3MB
or 10Mb Ethernet.
We hope to provide a (limited!) number of CPU's for running the
V-based agent code, for sites that want to experiment with the
multicasting facility. Contact the TF chairman (braden@isi.edu)
if you are interested.
Karen Lam at BBN is working on modifications to the 4.3BSD kernel
to allow a host to use Internet multicasting (Level 2, as defined
in RFC988). She has completed initial coding of the mods to 4.3BSD
and is just entering the compile/test/debug phase. She hopes to
have it working by the end of August (after the implementors'
workshop in Monterey), so it can be tested at Stanford against
Steve's multicast agents.
Eric Cooper at CMU has involved a number of students (from Alfred
Spector's Camelot project) in the design of possible multicast
applications. Their goal is to describe areas in the Camelot
design where unreliable (IP level) and reliable multicast (RPC
level) will be used.
BULK TRANSFER PROTOCOLS
Annette DeSchon (ISI) has been continuing work with Mark Lambert
of MIT on NETBLT protocol performance testing over the Wideband
8
network, between MIT and ISI. The fastest data rate that has been
observed so far is 228K bits per second. Based on their
measurements, the limiting factor appears to be the Ethernet
interface of the IBM PC/AT. Therefore, these tests will be
repeated using MIT's Symbolics LISP implementation as soon as the
appropriate hardware becomes available. ISI is also looking into
the feasibility of porting NETBLT to the Sun 3 workstation.
Gerd Beling (FGAN) and Bob Braden (ISI) did some thinking about
selective acknowledgments within the context of TCP. Existing TCP
implementations are generally tuned for short-delay paths, and
perform poorly over satellite circuits, for example. The
selective acknowledgment mechanism is being proposed as a
(compatible) extension which will allow a single TCP
implementation to be able to function efficiently over a much
broader range of delays than is currently possible.
TRANSACTION PROTOCOLS
Dave Cheriton is continuing work on his VTMP protocol at Stanford.
An experimental version of the V distributed system is running
with VMTP, loading programs and reading/writing files. A few bugs
remain. Dave will give a talk on VMTP at SIGCOMM, and is thinking
about an RFC which will contain a (revised) VMTP specification.
Eric Cooper at CMU is looking at the possible application of VMTP
for network message forwarders within Mach.
Eric Cooper and Karen White, while they were at Berkeley,
developed a remote procedure call mechanism for Berkeley Unix,
based upon the well-known Birrell&Nelson work. Their code was
integrated with the 4.2BSD kernel for efficiency. It would be
highly desirable for this code to be distributed as part of a
future BSD release, so other groups can experiment with RPC.
Jim Stevens, working on the SURAN packet radio effort, is
designing another Birrell&Nelson derivative called TTP, or
"Transaction Transport Protocol".
There will be a session at the upcoming TCP/IP Vendors' Workshop
on transaction protocols.
9
DATA STRUCTURING
Eric Cooper at CMU has been benchmarking the Courier
structured-data represention. He has timings and detailed
profiles for a standard example based upon a complex RFC822
header. One of Dave Clark's students is working on the same
example, using the structured-data representations in USP and
X.409.
Bob Braden
INTERNET ARCHITECTURE
1. Many of the INARCers are also INENGers and attended that
meeting in Ann Arbor this month. There are so many hot issues
buzzing with NSFNET and the explosive growth in connectivity
that it's hard to separate the near-term and far-term plates.
Several agenda ideas for the next INARC meeting emerged and
will be pursued.
2. In response to suggestions from the members of both task
forces, as well as the IAB, we are planning to coordinate the
next INARC and INENG meetings with the IAB meeting at RIACS in
October. The present plan is for the groups to meet together
for one day and separately for the other.
Dave Mills
INTERNET ENGINEERING
No report received.
INTEROPERABILITY
No report received.
PRIVACY
Plans for the next Privacy Task Force meeting, to be held on 11-12
September at University College London, were finalized. Work was
performed on generating a draft RFC entitled "Privacy Enhancement
for Internet Electronic Mail" for distribution to Task Force
members and review at the London meeting.
The Task Force welcomed a new member: Mike Padlipsky of
Mitre/Bedford.
John Linn
10
ROBUSTNESS AND SURVIVABILITY
No report received.
SCIENTIFIC COMPUTING
The first meeting of the Task Force on Scientific Networks was
held at NASA Ames Research Center, Research Institute for Advanced
Computer Science on June 26-27, 1986. Attending were:
Ron Bailey, NAS, NASA Ames (1st day)
Bob Braden, ISI
Bob Brown, RIACS, NASA Ames
Steve Casner, ISI
Alan Katz, ISI
Keith Lantz, Stanford
Jim Leighton, NMFECC
Barry Leiner, RIACS, NASA Ames
Creon Levit, NAS, NASA Ames (1st day)
Milo Medin, Bendix, NASA Ames
Mike Muuss, BRL
Ari Ollikainen, NAS/GE, NASA Ames
David Roode, BIONET
Peter Shames, STSI
1. The charter of the task force was reviewed and a general
discussion took place of the role of the internet technology
in supporting the scientific community.
2. The afternoon of the first day was spent discussing the
interactions between workstations and supercomputers and the
kinds of transactions required for scientific use over the
network.
3. The morning of the second day was spent on two applications
that require networking. These were structured graphics and
distributed ray-tracing.
4. The afternoon was spent discussing support of collaborative
research, particularly using multimedia conferencing.
The next meeting is scheduled for November 13-14 at Livermore. At
that time, we will identify some specific work topics for the task
force.
SECURITY
No report received.
11
TACTICAL INTERNET
No report received.
TESTING AND EVALUATION
No report received.
12