<NIS.NSF.NET> [IMR] IMR86-10.TXT
Westine [Page 1]
~
OCTOBER 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 distri-
bution.
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@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
---------------------------------------------------
VAX NETWORKING
In the month of October, work continued on the design and implemen-
tation of the Multicast Agent for 4.3 BSD Unix. The user-level
part of the Agent has been completed, and about half of the
kernel-level implementation has been written. Several Agent
specifications issues are still pending.
Westine [Page 1]
Internet Monthly Report October 1986
Karen Lam spent 20-22 October at Stanford University, testing and
debugging the 4.3 BSD Multicast Agent and IGMP implementations
against the Stanford software. On 23-24 October, Karen Lam
attended the Third Annual Berkeley Unix Workshop, held at the
University of California, Berkeley. Steve Deering, of Stanford
University, gave a presentation on Internet Multicast, paying some
attention to the 4.3 BSD IGMP implementation.
WIDEBAND NETWORK
On October 22 the Wideband Network provided the long-haul communi-
cation required for a multi-site demonstration of the Cronus Dis-
tributed Operating System. The demonstration was given in conjunc-
tion with the Rome Air Development Center's annual Distributed Sys-
tem Technology Exchange meeting. A Cronus application called the
C2 Internet Experiment was run on an integrated variety of computer
bases which were located in two geographical "clusters", one at
RADC and the other at BBN. Local communication within each cluster
was provided via Ethernet; long-haul communication between the
clusters was provided by the Wideband Network. Connectivity
between the Ethernets, the Wideband Net, and the ARPANET was pro-
vided by Butterfly Internet Gateways. After resolution of a
hardware problem with the RADC Internet Gateway the demonstration
worked flawlessly and was very well received.
Final testing of BSAT Release 2.1 was completed and the new
software was distributed to the Wideband sites on October 23. This
release contains many operational enhancements as well as a number
of bug fixes. The most significant of the enhancements are (1) a
"Mini-NOC" facility allowing control of any arbitrary set of remote
BSATs to be exercised from any local BSAT console, (2) a BSAT
built-in satellite channel bit error rate testing module, and (3) a
facility for fixed group (multicast) addressing. The Mini-NOC
facility reduces dependence on the the C/70 computer currently used
for monitoring and control of the Wideband sites. Item (2) allows
for non-invasive satellite channel BER measurement, i.e., measure-
ment that does not require the suspension of normal network opera-
tions. Item (3) will be used to support 3- (or more) way confer-
ences between Wideband sites.
A change in the Wideband Butterfly Internet Gateways was initiated
during the month in which the the affected gateways advertise a
lower cost route for the Wideband Net than for the ARPANET. This
has the effect of forcing certain LAN to LAN traffic that would
have previously traversed the ARPANET to traverse the Wideband net-
work, thereby alleviating some of the current ARPANET congestion
problems. The Wideband Net is not exhibiting any problems in sup-
porting the current level of re-routed traffic.
Westine [Page 2]
Internet Monthly Report October 1986
SATNET
As in past months, the SATNET continued to be stable. We began
looking at some of the channel 1 problems reported last month, but
will have to wait for the Linkabit spares if any repairs are
needed. Since channel 0 performance was fine, SATNET service to
the user sites was uninterrupted.
We realize that there have been interruptions in the overall ser-
vice provided by SATNET, largely due to problems with the new But-
terfly gateways. However, despite these problems, the SATNET
itself (SIMPs, PSP terminals, satellite equipment) has continued to
perform well. And as the gateways improve, the overall service
should continue to improve.
There were 5 scheduled outages during which the SATNET measurement
taskforce conducted some experiments. Also, service was inter-
rupted for about 24 hours on 10/6 and 10/7 when 2 phone lines
between the SATNET and the ARPAnet failed.
GATEWAYS
The major event of the month was the installation of new software
in the CSS Butterfly Gateway (ARPANET-Satnet). This version of the
software (Rel. 3.5) includes a number of bug fixes and minor
enhancements. The new software appears to much more robust and at
as this report was written had been up at CSS for a week with out
any crashes or restarts. This the new software was just installed
at UCL and RSRE and will be installed at the other Satnet sites
next.
As noted in more detail in Wideband network section of this report,
we reduced the routing cost advertised for the Wideband network to
have some LAN to LAN traffic be routed over the Wideband network
instead of the ARPANET.
Bob Hinden
ISI
---
General
Paul Mockapetris, attended the INARC/INENG Task Force meeting
October 15-17 and participated in the domain system planning.
Jon Postel and Bob Braden attended IAB meetings at RIACS
October 13-15. Jon Postel attended the "Berkeley Unix
Westine [Page 3]
Internet Monthly Report October 1986
Workshop" at UC Berkeley Oct 23-24. We welcomed three new
employees to ISI Division 2, Phil Dennis, Tom Hibbard and
Steve Farnworth.
Internet Concepts Project
Paul Mockapetris released a new version of the TOPS-20 domain
system JEEVES, which includes a resolver which uses aged round
trip times to optimize timeouts. New, higher performance
nameservers were installed at ISI and the NIC. Greg Finn con-
tinued studying issues involved in large scale routing.
Multimedia Conferencing Project
Brian Hung is working on a program to compose multimedia mes-
sages containing bitmap data on the IBM-PC AT. The goal is to
combine this program with the existing document scanning pro-
gram to allow multimedia messages containing bitmap data to be
composed and sent to another multimedia workstation.
Supercomputer and Workstation Communication Project
Bob Braden and Annette DeSchon have continued the the testing
of the NETBLT protocol with Mark Lambert, at MIT, over the
wideband network. In addition to the testing between two IBM
PCs running Mark's NETBLT implementation, we have been testing
Bob's Sun version of NETBLT against the IBM PC version running
at MIT. To support this testing Annette added more functions
to the SpyTool, which runs in the Xerox Development Environ-
ment on a Dandelion workstation.
Alan Katz attended the Supercomputer workshop at the San Diego
Supercomputer Center October 1 and 2.
MIT-LCS
-------
No internet related progress to report.
Lixia Zhang
NTA & NDRE
----------
No report received.
Westine [Page 4]
Internet Monthly Report October 1986
SRI
---
No report received.
UCL
---
1. The experiments on clock synchronisation have now finished,
though they are not complete. We have established the condi-
tions under which it is possible to synchronise clocks between
the US and the UK within acceptable limits. The very large
variance in round trip times requires careful filtering and
monitoring of the synchronisation data before adjustments are
made to the synchronised time. When we had the clocks syn-
chronised, we were able to measure the one-way delay between
UCL and a number of hosts in the DCN swamp. We found that the
outward delay time (UK->US) has a very large variance, whereas
the return delay time (US->UK) has a very low variance. This
has implications for the synchronisation process, which
assumes that the two one-way journey times are the same. The
variance is most probably another symptom of the overloading
of the ARPANET and its problems in supporting gateways. We
have allowed for the one-way delay variance in our filtering
technique. A paper is being prepared which describes our
experiments and results.
2. Some IP-level throughput tests were conducted on the UCL-
Goonhilly path during one of the SATNET reserved test periods.
Hence the path was essentially idle apart from the measurement
traffic. The measurement host was on a UCL ring, while the
echoer was (a) the UCL PDP-11 gateway, and (b) the Goonhilly
SIMP. When echoing 246-byte IP/ICMP packets off the UCL PDP
gateway, total throughput was 40 kbps (20k in each direction).
Beyond this rate, the UCL SAM started to drop packets because
its output queue to the gateway was growing too long; the SAM
and the PDP are connected by an LH/DH 1822. When echoing with
the same size packets off the Goonhilly SIMP, total throughput
was 17 kbps, with a 14% packet loss rate. Higher offered loads
resulted in up to 50% packet loss and frequent 'source quench'
messages from the PDP gateway. Keeping the packet loss rate
to less than 1% restricted the total throughput to 14 kbps,
i.e. 34% of the performance to the gateway. The UCL-Goonhilly
line currently runs the VDH protocol.
Peter Lloyd
Westine [Page 5]
Internet Monthly Report October 1986
UDEL
----
1. The UDEL1 fuzzball was joined on the UDEL Etherswamp by two
more fuzzballs, a radio clock and two IBM PCs. Lines are being
installed to various off-campus swamps as well. We are
currently reorganizing the entire UDEL swampland to install
subnetting and additional gateways for SURAnet, which is one
of the NSF regional nets.
2. The two DECNET fuzzballs and associated gear at Linkabit con-
tinue to paddle an awesome amount of traffic for NSFnet, FORD-
net, UMICHnet, UMDnet and their clients. Currently, sixteen
nets scattered all over the country splash at least some of
their packets across the DCnet swamp. The CMU gateway is now
functioning, but only in the direction towards the core, so
DCN-GATEWAY is still humping about a megapacket per week in
the other direction. Obviously, almost none of the routing
paths are symmetric.
3. I completed a preliminary report on the MSS routing study and
presented the results at an MSS meeting at Linkabit. I also
met with the Linkabit Dissimilar Gateway Protocol (DGP) pro-
ject staff and made several suggestions on the architecture
and protocols which might be developed.
4. My principal mailbox MILLS@D.ISI.EDU has been moved to a MIL-
NET host MILLS@A.ISI.EDU, which is mostly unreachable due net-
work and gateway congestion. I am requesting mail be sent
instead to MILLS@HUEY.UDEL.EDU until further notice.
5. I found and fixed a couple of bugs in the new Network Time
Protocol (NTP) daemon for the fuzzball. Clones are running now
on most fuzzballs, including those with radio clocks. Mike
Petry at U Maryland is operating a Unix 4.3bsd daemon with
filtering and deglitching code based on the fuzzball implemen-
tation. Those swamps without radio clocks, including the
NSFNET clients and UDEL fuzzies, now use NTP exclusively to
synchronize with radio timetellers.
6. I attended the IAB and task-force meetings at SRI, as well as
the NSF Network Program Advisory Group meeting at RIACS.
Dave Mills
Westine [Page 6]
Internet Monthly Report October 1986
TASK FORCE REPORTS
------------------
APPLICATIONS
No report received.
END-TO-END SERVICES
This report covers the 3 months August-October 1986.
The Task Force met September 3-4 at UCL, in conjunction with a
joint US/UK workshop on Protocols for Distributed Computing.
The 40 attendees from the UK were selected research people
actively concerned with network protocol development and
implementation. There was a lively discussion in addition to
the seven papers, ranging over topics including: protocols for
integrated service interfaces, high-performance transport pro-
tocols, the MACH and Camelot projects at CMU, Internet multi-
casting, the ANSA project at Cambridge, and transaction proto-
cols, especially REX (UK) and VMTP (US).
MULTICAST and HOST GROUPS
This work has become a cooperative effort between Stanford and
BBN. Karen Lam of BBN has completed the modifications to the
4.3BSD kernel to send and receive Internet multicasts. Her
code has been running several weeks on a couple of Stanford
machines.
Lam is now writing a multicast agent to run under 4.3BSD, hop-
ing to make an initial version available for others to test
before the end of 1986. The packet forwarding will be in the
kernel, while the code to maintain the host-group database
will execute in user mode. Meanwhile, Steve Deering is writ-
ing the specification for the protocol to be used between mul-
ticast agents.
Steve Deering attended the 3rd Annual Berkeley UNIX Workshop
October 23-24 and gave a talk on "Internet Multicast and
Berkeley UNIX".
Members of the Taskforce have started looking into applica-
tions of Internet multicasting, with special concern for its
impact on transport protocols. Applications under considera-
tion include resource location and distributed binding, e.g.
in MACH.
Westine [Page 7]
Internet Monthly Report October 1986
Discussions have been taking place with David Young of
Rockwell, who has been tasked under government contract to
develop a multicasting facility for SURAN. It is not yet
clear whether there will be any convergence between the SURAN
protocols developed independently by Rockwell and the main
body of Internet protocol research, but these discussions are
a hopeful sign.
TRANSACTION PROTOCOLS
Dave Cheriton continues to refine his VMTP protocol, and he is
working on a protocol specification RFC. Implementation of
VMTP under the V kernel is proceeding at Stanford, and work
has begun on a UNIX kernel implementation. Dave has done some
preliminary performance measurements on VMTP and plans to
write them up.
Cheriton's work on VMTP is also relevant to the DSAB (Distri-
buted Systems Architectures Board) and to Eric Cooper's work
at CMU on networking in MACH.
Jon Crowcroft and others at UCL are working on an ISO-style
directory service using Marshall Rose's ISODE (ISO Develope-
ment Environment), and in particular they are using the ISO
transaction protocol ROS for remote directory queries. ROS
provides "RPC" type semantics as well as asynchronous and no-
reply type communication.
The UCL Sequential Exchange Protocol (called SEP, for reasons
that only an Englishman could understand) has been completed,
and an RFC describing it is pending publication. A kernel
implementation for a SUN is available from UCL.
BULK TRANSFER PROTOCOLS
Experiments are continuing on NETBLT performance between MIT
and ISI over the Wideband Sattelite network. ISI ported Mark
Lambert' MIT version of NETBLT (for an IBM-PC) onto a SUN 3,
running in user mode and sending and receiving IP datagrams
using a RAW socket. Performance of the SUN version is
surprisingly good; between two SUN 3's on the ISI Ethernet, we
see up to 140 data packets/second ( 1.6Mbits/sec) with very
little packet loss. With a few lost packets, throughput drops
to 1.3 Mbits/sec. These tests are pure transport level, not
sending real data. Major objectives of the experiments now
include: understanding the observed 5-10% packet loss using
the Wideband net, pushing Wideband net throughput nearer to
the theoretical maximum, and developing timer algorithms which
can effectively adapt to both the Wideband net and the Ether-
net delays.
Westine [Page 8]
Internet Monthly Report October 1986
At the last meeting, some interest was expressed in working on
multicast bulk transfer protocols, which could be used for
dissemination of databases simultaneously to multiple sites.
We know about the Project Admiral work in this area; we would
like to hear from any other groups interested in this problem.
STRONG DATA TYPING
UCL is working on replicating the CMU measurements (ref. July
report) of the Courier structured data representation. They
hope to extend this to measure X.409 stubs (which are gen-
erated by parsing Abstract Syntax Notation (ASN.1) specs for
services) to compare these measurements in some meaningful way
with Courier and UCL RPC, and with Dave Clark's measurements
at MIT.
NETBIOS
At the TCP/IP Vendor Workshop, a vendor group was assembled to
create a standard Internet mapping for the NETBIOS interface.
An administrative mechanism has been created to give an offi-
cial stamp to the results of this or any similar vendor stan-
dardization effort: IAB task forces will perform this func-
tion.
In particular, the End-to-End Protocols task force was desig-
nated as the approval agency for the NETBIOS effort; Lorenzo
Aguilar of SRI is handling this job for the task force. NET-
BIOS is also interesting to us because it provides a natural
application for the Internet multicasting facility.
NETWORK FILE SYSTEMS
Members of the Internet community have expressed interest in
the definition of an Internet standard for network file access
protocols. (NB: We are talking about a protocol for ALL
HOSTS, not just UNIX systems!) The task force is considering
setting up a working party of people from the research and
vendor communities, with some interest and experience in this
area. If it is decided to proceed in this direction, an
announcement will be made sometime early next year, but anyone
interested could meanwhile drop a note to braden@isi.edu, out-
lining their goals and interests in this effort.
Bob Braden
Westine [Page 9]
Internet Monthly Report October 1986
INTERNET ARCHITECTURE
1. The INARC and INENG task forces held a joint meeting at SRI on
16 October. and separately on one other day. At the joint
meeting several issues of common interest were discussed,
including the evolution of EGP, transition to ISO and deploy-
ment of the domain name system. At the private INARC meeting
on 17 October ad-hoc presentations were heard from BBN, Xerox
and Mitre project leaders. Phill Gross will distribute
detailed minutes and presentation documents at a later date.
2. Considerable discussion was held in both the IAB and INARC on
the charter and agenda of the INARC, as well as the other task
forces. The discussion revealed widely separated views on the
basic organization and function of these groups. The IAB chair
has suggested the creation of two new groups. One of these
would be committed to a near-term technology maturing in about
five years and specific to the NSF networking community. The
other would be committed to long-range research beyond that
term. The INARC charter presently commits that group to
research in advancing technology without reference to specific
milestones and without obligation to specific system planning
or engineering. The results of discussion indicate that a)
INARC is not either of these new groups and b) an in-depth
review of the INARC charter, agenda and membership needs to be
conducted to determine whether and how the existing task force
fits in with the new structure.
3. The special-interest group formed to study issues involved
with campus networks, in particular the state and regional
networks associated with the NSF community, is engaged in
lively debate. Among the timely issues being discussed are
testing of the ad-hoc gateway complex now in place, how to
distribute the burgeoning traffic load over the ARPANET gate-
ways and how to connect the many regional and campus networks
operating with very different routing architectures and proto-
cols to the system.
Dave Mills
INTERNET ENGINEERING
No report received.
Westine [Page 10]
Internet Monthly Report October 1986
INTEROPERABILITY
No internet related progress to report.
Deborah Estrin
PRIVACY
Limited Privacy Task Force activity took place in the month of
October. The process of revising the draft Electronic Mail
Privacy Enhancement RFC in response to discussion at the Sep-
tember task force meeting began, as did the process of assem-
bling minutes from that meeting.
John Linn
ROBUSTNESS AND SURVIVABILITY
No report received.
SCIENTIFIC COMPUTING
No report received.
SECURITY
No report received.
TACTICAL INTERNET
No report received.
TESTING AND EVALUATION
No report received.
Westine [Page 11]