rfc889.txt

RFC 的详细文档！

Network Working Group                                   D.L.  Mills
Request for Comments:  889                              December 1983


                          Internet Delay Experiments

This memo reports on some measurement experiments and suggests some possible
improvements to the TCP retransmission timeout calculation.  This memo is
both a status report on the measurements and advice to implementers of TCP.

1.  Introduction

     This memorandum describes two series of experiments designed to explore
the transmission characteristics of the Internet system.  One series of
experiments was designed to determine the network delays with respect to
packet length, while the other was designed to assess the effectiveness of the
TCP retransmission-timeout algorithm specified in the standards documents.
Both sets of experiments were conducted during the October - November 1983
time frame and used many hosts distributed throughout the Internet system.

     The objectives of these experiments were first to accumulate experimental
data on actual network paths that could be used as a benchmark of Internet
system performance, and second to apply these data to refine individual TCP
implementations and improve their performance.

     The experiments were done using a specially instrumented measurement host
called a Fuzzball, which consists of an LSI-11 running IP/TCP and various
application-layer protocols including TELNET, FTP and SMTP mail.  Among the
various measurement packages is the original PING (Packet InterNet Groper)
program used over the last six years for numerous tests and measurements of
the Internet system and its client nets.  This program contains facilities to
send various kinds of probe packets, including ICMP Echo messages, process the
reply and record elapsed times and other information in a data file, as well
as produce real-time snapshot histograms and traces.

     Following an experiment run, the data collected in the file were reduced
by another set of programs and plotted on a Peritek bit-map display with color
monitor.  The plots have been found invaluable in the indentification and
understanding of the causes of netword glitches and other "zoo" phenomena.
Finally, summary data were extracted and presented in this memorandum.  The
raw data files, including bit-map image files of the various plots, are
available to other experimenters upon request.

     The Fuzzballs and their local-net architecture, called DCN, have about
two-dozen clones scattered worldwide, including one (DCN1) at the Linkabit
Corporation offices in McLean, Virginia, and another at the Norwegian
Telecommunications Adminstration (NTA) near Oslo, Norway.  The DCN1 Fuzzball
is connected to the ARPANET at the Mitre IMP by means of 1822 Error Control
Units operating over a 56-Kbps line.  The NTA Fuzzball is connected to the
NTARE Gateway by an 1822 interface and then via VDH/HAP operating over a
9.6-Kbps line to SATNET at the Tanum (Sweden) SIMP.  For most experiments
described below, these details of the local connectivity can be ignored, since
only relatively small delays are involved.




Internet Delay Experiments                                              Page 2
D.L. Mills


     The remote test hosts were selected to represent canonical paths in the
Internet system and were scattered all over the world.  They included some on
the ARPANET, MILNET, MINET, SATNET, TELENET and numerous local nets reachable
via these long-haul nets.  As an example of the richness of the Internet
system connectivity and the experimental data base, data are included for
three different paths from the ARPANET-based measurement host to London hosts,
two via different satellite links and one via an undersea cable.

2.  Packet Length Versus Delay

     This set of experiments was designed to determine whether delays across
the Internet are significantly influenced by packet length.  In cases where
the intrinsic propagation delays are high relative to the time to transmit an
individual packet, one would expect that delays would not be strongly affected
by packet length.  This is the case with satellite nets, including SATNET and
WIDEBAND, but also with terrestrial nets where the degree of traffic
aggregation is high, so that the measured traffic is a small proportion of the
total traffic on the path.  However, in cases where the intrinsic propagation
delays are low and the measured traffic represents the bulk of the traffic on
the path, quite the opposite would be expected.

     The objective of the experiments was to assess the degree to which TCP
performance could be improved by refining the retransmission-timeout algorithm
to include a dependency on packet length.  Another objective was to determine
the nature of the delay characteristic versus packet length on tandem paths
spanning networks of widely varying architectures, including local-nets,
terrestrial long-haul nets and satellite nets.

2.1.  Experiment Design

     There were two sets of experiments to measure delays as a function of
packet length.  One of these was based at DCN1, while the other was based at
NTA.  All experiments used ICMP Echo/Reply messages with embedded timestamps.
A cycle consisted of sending an ICMP Echo message of specified length, waiting
for the corresponding ICMP Reply message to come back and recording the
elapsed time (normalized to one-way delay).  An experiment run, resulting in
one line of the table below, consisted of 512 of these volleys.

     The length of each ICMP message was determined by a random-number
generator uniformly distributed between zero and 256.  Lengths less than 40
were rounded up to 40, which is the minimum datagram size for an ICMP message
containing timestamps and just happens to also be the minimum TCP segment
size.  The maximum length was chosen to avoid complications due to
fragmentation and reassembly, since ICMP messages are not ordinarily
fragmented or reassembled by the gateways.

     The data collected were first plotted as a scatter diagram on a color
bit-map display.  For all paths involving the ARPANET, this immediately
revealed two distinct characteristics, one for short (single-packet) messages
less than 126 octets in length and the other for long (multi-packet) messages



Internet Delay Experiments                                              Page 3
D.L. Mills


longer than this.  Linear regression lines were then fitted to each
characteristic with the results shown in the following table.  (Only one
characteristic was assumed for ARPANET-exclusive paths.) The table shows for
each host the delays, in milliseconds, for each type of message along with a
rate computed on the basis of these delays.  The "Host ID" column designates
the host at the remote end of the path, with a letter suffix used when
necessary to identify a particular run.














































Internet Delay Experiments                                              Page 4
D.L. Mills


Host    Single-packet   Rate    Multi-packet    Rate    Comments
ID      40      125     (bps)   125     256     (bps)
---------------------------------------------------------------------------
DCN1 to nearby local-net hosts (calibration)
DCN5    9                               13      366422  DMA 1822
DCN8    14                              20      268017  Ethernet
IMP17   22                              60      45228   56K 1822/ECU
FORD1   93                              274     9540    9600 DDCMP base
UMD1    102                             473     4663    4800 synch
DCN6    188                             550     4782    4800 DDCMP
FACC    243                             770     3282    9600/4800 DDCMP
FOE     608                             1917    1320    9600/14.4K stat mux

DCN1 to ARPANET hosts and local nets
MILARP  61      105     15358   133     171     27769   MILNET gateway
ISID-L  166     263     6989    403     472     15029   low-traffic period
SCORE   184     318     5088    541     608     15745   low-traffic period
RVAX    231     398     4061    651     740     11781   Purdue local net
AJAX    322     578     2664    944     1081    7681    MIT local net
ISID-H  333     520     3643    715     889     6029    high-traffic period
BERK    336     967     1078    1188    1403    4879    UC Berkeley
WASH    498     776     2441    1256    1348    11379   U Washington

DCN1 to MILNET/MINET hosts and local nets
ISIA-L  460     563     6633    1049    1140    11489   low-traffic period
ISIA-H  564     841     2447    1275    1635    2910    high-traffic period
BRL     560     973     1645    1605    1825    4768    BRL local net
LON     585     835     2724    1775    1998    4696    MINET host (London)
HAWAII  679     980     2257    1817    1931    9238    a long way off
OFFICE3 762     1249    1396    2283    2414    8004    heavily loaded host
KOREA   897     1294    1712    2717    2770    19652   a long, long way off

DCN1 to TELENET hosts via ARPANET
RICE    1456    2358    754     3086    3543    2297    via VAN gateway

DCN1 to SATNET hosts and local nets via ARPANET
UCL     1089    1240    4514    1426    1548    8558    UCL zoo 
NTA-L   1132    1417    2382    1524    1838    3339    low-traffic period
NTA-H   1247    1504    2640    1681    1811    8078    high-traffic period

NTA to SATNET hosts
TANUM   107                             368     6625    9600 bps Tanum line
ETAM    964                             1274    5576    Etam channel echo
GOONY   972                             1256    6082    Goonhilly channel echo









Internet Delay Experiments                                              Page 5
D.L. Mills


2.2 Analysis of Results

     The data clearly show a strong correlation between delay and length, with
the longest packets showing delays two to three times the shortest.  On paths
via ARPANET clones the delay characteristic shows a stonger correlation with
length for single-packet messages than for multi-packet messages, which is
consistent with a design which favors low delays for short messages and high
throughputs for longer ones.

     Most of the runs were made during off-peak hours.  In the few cases where
runs were made for a particular host during both on-peak and off-peak hours,
comparison shows a greater dependency on packet length than on traffic shift.

     TCP implementors should be advised that some dependency on packet length
may have to be built into the retransmission-timeout estimation algorithm to
insure good performance over lossy nets like SATNET.  They should also be
advised that some Internet paths may require stupendous timeout intervals
ranging to many seconds for the net alone, not to mention additional delays on
host-system queues.

     I call to your attention the fact that the delays (at least for the
larger packets) from ARPANET hosts (e.g.  DCN1) to MILNET hosts (e.g.  ISIA)
are in the same ballpark as the delays to SATNET hosts (e.g.  UCL)!  I have
also observed that the packet-loss rates on the MILNET path are at present not
neglible (18 in 512 for ISIA-2).  Presumably, the loss is in the gateways;
however, there may well be a host or two out there swamping the gateways with
retransmitted data and which have a funny idea of the "normal" timeout
interval.  The recent discovery of a bug in the TOPS-20 TCP implementation,
where spurious ACKs were generated at an alarming rate, would seem to confirm
that suspicion.

3.  Retransmission-Timeout Algorithm

     One of the basic features of TCP which allow it to be used on paths
spanning many nets of widely varying delay and packet-loss characteristics is
the retranansmission-timeout algorithm, sometimes known as the "RSRE
Algorithm" for the original designers.  The algorithm operates by recording
the time and initial sequence number when a segment is transmitted, then
computing the elapsed time for that sequence number to be acknowledged.  There
are various degrees of sophistication in the implementation of the algorithm,
ranging from allowing only one such computation to be in progress at a time to
allowing one for each segment outstanding at a time on the connection.

     The retransmission-timeout algorithm is basically an estimation process.
It maintains an extimate of the current roundtrip delay time and updates it as
new delay samples are computed.  The algorithm smooths these samples and then
establishes a timeout, which if exceeded causes a retransmission.  The
selection of the parameters of this algorithm are vitally important in order
to provide effective data transmission and avoid abuse of the Internet system
by excessive retransmissions.  I have long been suspicious of the parameters



Internet Delay Experiments                                              Page 6
D.L. Mills


suggested in the specification and used in some implementations, especially in
cases involving long-delay paths involving lossy nets.  The experiment was
designed to simulate the operation of the algorithm using data collected from
real paths involving some pretty leaky Internet plumbing.

3.1.  Experiment Design

     The experiment data base was constructed of well over a hundred runs
using ICMP Echo/Reply messages bounced off hosts scattered all over the world.
Most runs, including all those summarized here, consisted of 512 echo/reply
cycles lasting from several seconds to twenty minutes or so.  Other runs
designed to detect network glitches lasted several hours.  Some runs used
packets of constant length, while others used different lengths distributed
from 40 to 256 octets.  The maximum length was chosen to avoid complications
fragmented or reassembled by the gateways.

     The object of the experiment was to simulate the packet delay
distribution seen by TCP over the paths measured.  Only the network delay is
of interest here, not the queueing delays within the hosts themselves, which
can be considerable.  Also, only a single packet was allowed in flight, so
that stress on the network itself was minimal.  Some tests were conducted
during busy periods of network activity, while others were conducted during
quiet hours.

     The 512 data points collected during each run were processed by a program
which plotted on a color bit-map display each data point (x,y), where x
represents the time since initiation of the experiment the and y the measured
delay, normalized to the one-way delay.  Then, the simulated
retransmission-timeout algorithm was run on these data and its computed
timeout plotted in the same way.  The display immediately reveals how the
algorithm behaves in the face of varying traffic loads, network glitches, lost
packets and superfluous retransmissions.

     Each experiment run also produced summary statistics, which are
summarized in the table below.  Each line includes the Host ID, which
identifies the run.  The suffix -1 indicates 40-octet packets, -2 indicates
256-octet packets and no suffix indicates uniformly distributed lengths
between 40 and 256.  The Lost Packets columns refer to instances when no ICMP
Reply message was received for thirty seconds after transmission of the ICMP
Echo message, indicating probable loss of one or both messages.  The RTX
Packets columns refer to instances when the computed timeout is less than the
measured delay, which would result in a superfluous retransmission.  For each
of these two types of packets the  column indicates the number of instances
and the Time column indicates the total accumulated time required for the
recovery action.

     For reference purposes, the Mean column indicates the computed mean delay
of the echo/reply cycles, excluding those cycles involving packet loss, while
the CoV column indicates the coefficient of variation.  Finally, the Eff




Internet Delay Experiments                                              Page 7