<NIS.NSF.NET> [IMR] IMR84-12.TXT
DECEMBER 1984
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 contractors in the ARPA Internet
Research Program.
This report is for research use only, and is not for public
distribution.
Each 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@USC-ISIF.
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
---------------------------------------------------
TOPS-20 TCP/IP
Jil Westcott and Charles Lynn attended a meeting at DEC
Marlboro to discuss schedules for including the current BBN
TOPS20 Internet code into release 6 of the TOPS20 monitor.
Merging the code continues to progress, but was slower than
anticipated due to investigation of bug reports (the TCP has
been caught exceeding the window) and the problem noted below.
Charles Lynn attended a meeting at the DDN PMO to discuss
1
delay problems between SAC/Ft. Bragg/DARPA and ISI and to plan
how to determine what is happening in each of component of the
internet when an user is in a long delay situation.
Continuous monitoring at ISI has been expanded to collect
connection data which can be examined after a problem is
reported. Hopefully, the resulting ocean of bits will contain
clues to the problem or indicate that the wrong data is being
collected.
GATEWAYS
The design of the Butterfly gateway continues to progress. We
are now at the stage where we will begin to write some code to
try out some of our design ideas.
The new Host Monitoring Protocol (HMP) RFC was released. We
are working on a revision of the EGP specification to handle
"stub" gateways which we expect to release in early January.
Plans for the Gateway SIG to be held at ISI the last two days
of Feburary are progressing.
SATNET
The C/30 to replace the Honeywell 316 Satellite IMP at Tanum
was shipped to Norway and is currently in Oslo. BBN Field
Service will install the machine as soon as it has been
shipped from there to Tanum, Sweden.
WIDEBAND SATELLITE NETWORK
The Wideband Network operated with three sites on the
satellite channel throughout most of December. A PSAT
software bug, limiting the number of sites on the channel, was
uncovered while attempting to bring up the Ft. Monmouth site
during December. It is hoped that this problem will be
resolved in early January. The ISI site has been operating
with a new ESI-A stably since December 7. The old ESI from
ISI was refurbished by Linkabit during the month and has been
shipped to SRI. SRI is expected to be returned to full
operational status during early January.
VAX UNIX TCP/IP
This month, work progressed on the port of the BBN TCP/IP to
4.2 BSD. The code compiles and link-edits, and we have begun
testing. Some of the questions resolved involved the routing
algorithms (i.e., whether we should use Berkeley's or retain
our own), and the larger question of how much of the interface
code between the TCP/IP code and the socket (user interface)
to use.
2
Bob Hinden
ISI
---
INTERNET PROJECT
GENERAL: Jon Postel and Joyce Reynolds visited the NIC for
discussions on the Domain Plan and to coordinate on a
number of protocol administrative matters. Joyce Reynolds
gave a seminar on CSNET.
DOMAINS: Paul Mockapetris and Ruth Brungardt worked on the
domain system implementation.
MULTIMEDIA MAIL: Alan Katz worked on implementing ST and
IP.ST protocols on the PERQ in order to communicate with
the PVT voice terminals for multimedia mail. Joyce
Reynolds made preparations for Multimedia Mail exchanges
with BBN.
TEST TOOLS: Greg Finn worked on the testing of the 1108
Lisp implementation of IP, ICMP, and Address Resolution
protocols continues. Discussion with Xerox concerning
interface to higher level protocols (TCP) is resulting in
some changes. Annette DeSchon has been implementing an
Address Resolution Protocol to run in the Mesa Development
Environment on the XEROX 8010. David Smallberg found a
snag in the implementation of the UDP little servers.
FILTER INTERFACE: Paul Mockapetris worked on the TCP
measurement code.
WIDEBAND PROJECT
No internet-related progress to report.
COMPUTER CENTER
Efforts were made to help isolate problems of slow response
between TACs and Tops-20s which had been observed on ISI
systems. A meeting was held at DDN/PMO with
representitives from BBN, ISI, DDN, and SAC to develop
plans for increased monitoring in the TAC, IMP, and TOPS-20
that could be activated during periods of poor response.
Tentative plans for an uncontrolled message experiment were
also made. In cooperation with Peter Ditmars, two tests
were run between ISIE and SAC-TAC. Tops-20 trace
information was collected and given to Peter. In
cooperation with Charles Lynn, the Tops-20 network
3
monitoring code was expanded, and a bug was found and fixed
that had been preventing monitoring datagrams from being
sent in reply to probes. Directories were set up on ISIA
and ISIE to hold collected data. Work to fit further
additions to the monitoring code into ISI Tops-20 is near
completion.
Joel Goldberger
LINCOLN LAB
-----------
A command has been added to the IP/ST gateway control and
monitoring capability that allows an ICMP echo request probe
to be sent to an arbitrary internet address. The probe may be
source-routed by typing additional IP addresses up to the
limit posed by the size of the option field in the IP header.
When the echo response is returned, a line is typed on the
monitor screen. Broadcast or group addresses may be used to
get multiple responses from sets of hosts or gateways.
(Interesting flurries of activity can be caused by such
addresses within the source route.)
The IP/ST gateways now send the current date and time across
the WB SATNET in the "please join my group" messages that they
send to each other. When a gateway receives such a message it
checks to see if its operating system had been brought up
without anyone entering the date and time. If such had been
the case, the gateway program will set the operating system
time as well as its own concept of time that it uses for
logging events of interest.
Time setting is the latest of a series of features that have
been added to speed up and simplify the process of bringing up
the IP/ST gateways in the field. Pushing the boot switch on
the front panel of the PDP 11/44 is now all that is needed to
bring up a gateway. During the past several months Steve
Casner of ISI has made a number of changes to the EPOS
operating system to support this capability and to provide us
with full access to virtual memory in the 11/44. We are
grateful for his efforts.
Jim Forgie
LINKABIT
--------
1. We packed up the fuzzballs for the move to new offices in
Vienna, Virginia. Our local-net zoo will be isolated pending
4
reconnection of our ARPANET access line, expected before 19
January.
2. With Jon's kind help, RFC889, a summary of my Internet
delay experiments reported last month, RFC891, a specification
of the protocols mumbled in our local-net zoo, and RFC892, a
copy of the ISO Transport Protoocl Specification, were
distributed.
3. Our TCP experiments continued with the rebuilding of the
buffering and packetization algorithms. The result of this was
a worthwhile increase in efficiency along with a reduction in
buffering requirments. The retransmission algorithm was
further refined to improve its behavior in cases of large
delay means and variances. Extensive tests with the new
trace/trap facilities over several ornery paths including
SATNET showed the changes are working well.
4. The TCP testing program continued to catch bugs previously
obscured by performance shortfalls. The TOPS-20 "runaway ACK"
bug continues to cause severe bleeding in our ARPANET gateway
and presumably others as well. At least one host on the
ARPANET continues to send jumbograms larger than 576 octets
without prior agreement, leading to busted 1822 packets. A
raunchy bug was discovered in which the TOPS-20 TCP sender
overran the reciever window, improperly repacketize the
retransmission and finally broke the connection. The client
(SMTP) then tried to re-send using 50-octet segments. Sound
familiar?
5. While cleaning up some old timekeeping business, I
discovered the local power grid is even less stable than I
thought. relative time slips exceeding 8 milliseconds per
second were observed over the Christmas break, well in excess
of anything measured before. I retuned the time-
synchronization parameters again to allow for these
excursions. We bought one of the new Heath WWV-synchronized
radio clocks, which are relatively inexpensive. Initial tests
show that these clocks can show errors up to a large part of a
second.
6. Our EGP gateway continues to track the routing information
from the BBN test gateway; however, on a number of occasions
the BBN test gateway itself apparently became isolated so that
we lost connectivity. Our new packet-trace/trap facility found
an alarming incidence of misrouted ARPANET packets and
resulting gushers of redirects. These spasms, almost certainly
due to flapping gateways and nets somewhere in the ooze, often
persisted up to two minutes and on some occasions to twenty
minutes. With so many GGP gateways now on the ARPANET, spasms
5
of this magnitude must be expected; however, the cost has
become very serious.
Dave Mills
MIT-LCS
-------
C Gateway
Management of exterior net routing information in MIT's
internal gateways is being changed to maintain a cache of
exterior nets with the best next hop to the net. The
immediate motivation for this change is that we need to reach
a B address net via an internal MIT gateway; currently all
packets destined to B net addresses are sent to one MIT
gateway which connects MIT with the ARPANET. For the moment
the cache will be managed statically. When MIT's interior
gateway protocol is designed, it will be used to initialize
and update each internal gateway's cache of routes to exterior
nets. This protocol will also be used to dynamically maintain
all MIT gateways' subnet routing tables. The software has been
designed to allow easy configuration of gateways to use 1)
static routing, 2) interior gateway protocol, or 3) exterior
and interior gateway protocols. Design and implementation of
the software are almost complete; debugging should start soon.
This month we noticed that 576 byte internet packets were not
getting from hosts on the ARPANET to hosts on our 10Mbit
ethernet. It turned out that three of our gateway
configurations were not handling 576 byte packets properly.
The ARPANET to PRONET gateway zeroed out the last byte of data
in a 576 byte packet; this appeared to happen somewhere in the
arpanet serial line input code. We added four trailer bytes
to the ARPANET device specification (so that reads from the
ARPANET could handle extra trailer/pad bytes) and the problem
went away. The Bridge Communications gateway had a different
problem. It was forwarding the entire packet, encapsulated in
the header and trailer received from the 10Mbit ethernet, out
onto the PRONET. One host was transmitting two extra bytes
onto the ethernet with its internet packets which caused 576
byte internet packets to become 578 byte packets (plus PRONET
header) when forwarded onto the PRONET. The PRONET drivers
are not prepared to receive such large packets and discarded
them. The Bridge Communications gateway was changed to only
forward the contents of the internet packet between nets. The
third configuation, whose 576 byte packet forwarding problem
we are still studying, is the LSI 11/03 10Mbit to PRONET
gateway. We notice that the last four bytes of data in a 576
6
byte internet packet are trashed on input from the 10Mbit
ethernet.
In late November we created, for the first time, a
configuration of the CGW that contains both a 10MB and a 3MB
Ethernet interface. After fixing a few name conflicts, the
configuration was released to Stanford where it ran without
any problems. The Stanford gateway, Golden Gate, now contains
three network interfaces and it still has room for 17 packet
buffers. Support for EGP is not included in the Golden Gate
configuration.
IBM Personal Computer
Since MS-DOS (the PC's operating system) does not support
multiple processes running concurrently, the user telnet was
modified to include server TFTP. This provides a more
convenient way to transfer many files to the PC. The TFTP
server may be turned on or off; the telnet user is notified
whenever a TFTP request is received and is asked to verify the
request before it is accepted. This telnet/TFTP scheme is
uncovering some naming problems. Most PCs on the net do not
have their names listed with the various name servers so some
form of numeric address must be used to specify which PC a
file should be tranferred to. But there is no common numeric
address format across machines, some do not even accept
numeric host names, so the user is left with annoying naming
inconsistencies.
The timeout strategy in TFTP was improved and the program now
works well with a range of networks, from a 300 baud serial
line to a 10Mb ethernet. It formerly would almost always
timeout on a 1200 baud line. A number of bugs were corrected
in the area of timers, as well.
A user TCP/Supdup was brought up on the PC; this program also
has a built-in TFTP. Unfortunately, there are only a few
machines on the net which run server Supdup.
The ethernet driver was modified to allow the interrupt vector
location it uses to be modified by individual users after the
program has been loaded. This change means that the ethernet
driver can run on an XT; the hard disk controller on an XT
uses the interrupt vector location the ethernet driver had
been using. Several bugs in the way the ethernet driver
decided whether a packet was valid were corrected, and a
change was made to the Address Resolution Protocol
implementation so that Internet address - ethernet address
binding may be changed dynamically.
7
Liza Martin
NTA
---
No report received.
SRI
---
1. Development of the Robustness II software is continuing.
The conversion of existing utilities to the new prom format
should be completed soon. In addition, a new file manager
utility is under development. Utilities for floppy disk and
bubble copy operations were updated.
2. Development of the virtual memory enhancements to the MOS
operating system is proceeding. A technical note covering the
various technical options and approaches has been drafted and
is being circulated internally for review.
3. Three papers covering the naming, addressing, routing and
component cooperation aspects of the problems in accommodating
internetwork dynamics have been written and are undergoing
internal review.
Jim Mathis
SRI-NIC
-------
No report received.
UCL
---
1. Connectivity via Satnet has been quite reasonable this
month. The VAN path continues to have reliability problems
with quite a number of calls being refused.
2. Some simple measurments done by Robert Cole using the VAN
path indicated a maximum throughput of about 1100 bps and an
average UK/US delay of about 0.85 second.
Peter Lloyd
8
-------