TCP-IP FAQ

George V. Neville-Neil courtesy. Published with permission.

Archive-name:tcp-ip/FAQ
Last-modified:  1996/4/1

Internet Protocol Frequently Asked Questions

Maintained by: George V. Neville-Neil (gnn@wrs.com)
Contributions from:
Ran Atkinson
Mark Bergman
Stephane Bortzmeyer
Rodney Brown
Dr. Charles E. Campbell Jr.
Phill Conrad 
Alan Cox
Rick Jones
Jon Kay 
Jay Kreibrich
William Manning
Barry Margolin 
Jim Muchow
Subu Rama
W. Richard Stevens 
 
Version 3.3


************************************************************************

	The following is a list of Frequently Asked Questions, and
their answers, for people interested in the Internet Protocols,
including TCP, UDP, ICMP and others.  Please send all additions,
corrections, complaints and kudos to the above address.  This FAQ will
be posted on or about the first of every month.

	This FAQ is available for anonymous ftp from :
ftp.netcom.com:/pub/gnn/tcp-ip.faq .  You may get it from my home page at
ftp://ftp.netcom.com/pub/gnn/gnn.html
	You can read the FAQ in HTMl format on Netcom or from the mirror 
site http://web.cnam.fr/Network/TCP-IP/tcp-ip.html

************************************************************************

Table of Contents:
Glossary
1) Are there any good books on IP?
2) Where can I find example source code for TCP/UDP/IP?
3) Are there any public domain programs to check the performance of an
IP link? 
4) Where do I find RFCs?
5) How can I detect that the other end of a TCP connection has
crashed?  Can I use "keepalives" for this?
6) Can the keepalive timeouts be configured?
7) Can I set up a gateway to the Internet that translates IP
addresses, so that I don't have to change all our internal addresses 
to an official network? 
8) Are there object-oriented network programming tools?
9) What other FAQs are related to this one?
10) What newsgroups contain information on networks/protocols?
11) Van Jacobson explains TCP congestion avoidance.
12) Can I use a single bit subnet?

Glossary:

I felt this should be first given the plethora of acronyms used in the
rest of this FAQ.

IP: Internet Protocol.  The lowest layer protocol defined in TCP/IP.
This is the base layer on which all other protocols mentioned herein
are built.  IP is often referred to as TCP/IP as well.

UDP: User Datagram Protocol.  This is a connectionless protocol built
on top of IP.  It does not provide any guarantees on the ordering or
delivery of messages.  This protocol is layered on top of IP.

TCP: Transmission Control Protocol.  TCP is a connection oriented
protocol that guarantees that messages are delivered in the order in
which they were sent and that all messages are delivered.  If a TCP
connection cannot deliver a message it closes the connection and
informs the entity that created it.  This protocol is layered on top
of IP.

ICMP:  Internet Control Message Protocol.  ICMP is used for
diagnostics in the network.  The Unix program, ping, uses ICMP
messages to detect the status of other hosts in the net.  ICMP
messages can either be queries (in the case of ping) or error reports,
such as when a network is unreachable.

RFC: Request For Comment.  RFCs are documents that define the
protocols used in the IP Internet.  Some are only suggestions, some
are even jokes, and others are published standards.  Several sites in
the Internet store RFCs and make them available for anonymous ftp.

SLIP:  Serial Line IP.  An implementation of IP for use over a serial
link (modem).  CSLIP is an optimized (compressed) version of SLIP that
gives better throughput.

Bandwidth:  The amount of data that can be pushed through a link in
unit time.  Usually measured in bits or bytes per second.

Latency:  The amount of time that a message spends in a network going
from point A to point B.

Jitter:  The effect seen when latency is not a constant.  That is, if
messages experience a different latencies between two points in a
network.

RPC:  Remote Procedure Call.  RPC is a method of making network access
to resource transparent to the application programmer by supplying a
"stub" routine that is called in the same way as a regular procedure
call.  The stub actually performs the call across the network to
another computer.

Marshalling:  The process of taking arbitrary data (characters,
integers, structures) and packing them up for transmission across a
network.

MBONE: A virtual network that is a Multicast backBONE.  It is still a
research prototype, but it extends through most of the core of the
Internet (including North America, Europe, and Australia).  It uses IP
Multicasting which is defined in RFC-1112.  An MBONE FAQ is available
via anonymous ftp from: ftp.isi.edu" There are frequent broadcasts of
multimedia programs (audio and low bandwidth video) over the MBONE.
Though the MBONE is used for mutlicasting, the long haul parts of the
MBONE use point-to-point connections through unicast tunnels to
connect the various multicast networks worldwide.


1) Are there any good books on IP?

A) Yes.  Please see the following:

Internetworking with TCP/IP Volume I
(Principles, Protocols, and Architecture)
Douglas E. Comer
Prentice Hall 1991 ISBN 0-13-468505-9

This volume covers all of the protocols, including IP, UDP, TCP, and
the gateway protocols.  It also includes discussions of higher level
protocols such as FTP, TELNET, and NFS.

Internetworking with TCP/IP Volume II
(Design, Implementation, and Internals)
Douglas E. Comer / David L. Stevens
Prentice Hall 1991  ISBN 0-13-472242-6

Discusses the implementation of the protocols and gives numerous code
examples.

Internetworking with TCP/IP Volume III (BSD Socket Version)
(Client - Server Programming and Applications)
Douglas E. Comer / David L. Stevens
Prentice Hall 1993  ISBN 0-13-474222-2

This book discusses programming applications that use the internet
protocols.  It includes examples of telnet, ftp clients and servers.
Discusses RPC and XDR at length.

TCP/IP Illustrated, Volume 1: The Protocols, 
W. Richard Stevens
(c) Addison-Wesley, 1994  ISBN 0-201-63346-9

An excellent introduction to the entire TCP/IP protocol suite,
covering all the major protocols, plus several important applications.

"TCP/IP Illustrated, Volume 2: The Implementation",
by Gary R. Wright and W. Richard Stevens
(c) Addison-Wesley, 1995
ISBN 0-201-63354-X

This is a complete, and lenthy, discussion of the internals of TCP/IP
based on the Net/2 release of BSD.

Unix Network Programming
W. Richard Stevens
Prentice Hall 1990  ISBN 0-13-949876

An excellent introduction to network programming under Unix.

The Design and Implementation of the 4.3 BSD Operating System
Samuel J. Leffler, Marshall Kirk McKusick, Michael J. Karels, John S.
Quarterman 
Addison-Wesley 1989  ISBN 0-201-06196-1

Though this book is a reference for the entire operating system, the
eleventh and twelfth chapters completely explain how the networking
protocols are implemented in the kernel.

Stevens, W. Richard, Unix Network Programming.  1990, Prentice-Hall.

An excellent introduction to network programming under Unix.   Widely
cited on the Usenet bulliten boards as the "best place to start" if you
want to actually learn how to write Unix programs that communicate over
a network.  

Rago, Steven A.  Unix System V. Network Programming.  1993, Addison-Wesley.

A book that covers the same kinds of topics as W. Richard Stevens Unix
Network Programming, but is more specific to Unix System V Release 4
(SVR4), and so perhaps is more useful and up to date if you are
working specifically with that implementation.  (Stevens book covers
Unix System V release 3.x).  There is a much more extensive coverage
of Streams in Rago's book; 4 chapters, where Stevens only provides a
couple of subsections.  The design project at the end of the book is
an implementation of SLIP.


2)  Where can I find example source code for TCP/UDP/IP?

A)  Code from the Internetworking with TCP/IP Volume III is available
for anonymous ftp from:

arthur.cs.purdue.edu:/pub/dls

Code used in the Net-2 version of Berkeley Unix is available for
anonymous ftp from:

ftp.uu.net:systems/unix/bsd-sources/sys/netinet 

and

gatekeeper.dec.com:/pub/BSD/net2/sys/netinet

Code from Richard Steven's book is available on:
ftp.uu.net:/published/books/stevens.*

Example source code and libraries to make coding quicker is available
in the Simple Sockets Library written at NASA.  The Simple Sockets
Library makes sockets easy to use!  And, it comes as source code.  It
has been tested on: Unix (SGI, DecStation, AIX, Sun 3, Sparcstation;
version 2.02+: Solaris 2.1, SCO), VMS, and MSDOS (client only since
there's no background there).  It is provided in source code form, of
course, and sits atop Berkeley sockets and tcp/ip.

You can order the "Simple Sockets Library" from

                           Austin Code Works
                          11100 Leafwood Lane
                       Austin, TX 78750-3464 USA
                         Phone (512) 258-0785
                 
Ask for the "SSL - The Simple Sockets Library".  Last I checked, they
were asking $20 US for it.


For DOS there is WATTCP.ZIP (numerous sites): 

WATTCP is a DOS TCP/IP stack derived from the NCSA Telnet program and
much enhanced. It comes with some example programs and complete source
code. The interface isn't BSD sockets but is well suited to PC type
work. It is also written so that it can be used and memory
allocation).

3)  Are there any public domain programs to check the performance of
an IP link?

A)  

TTCP:  Available for anonymous ftp from....

wuarchive.wustl.edu:/graphics/graphics/mirrors/sgi.com/sgi/src/ttcp

On ftp.sgi.com are netperf (from Rick Jones at HP) and nettest
(from Dave Borman at Cray).  ttcp is also availabel at ftp.sgi.com.

You can get to the NetPerf home page via:

http://www.cup.hp.com/netperf/NetperfPage.html


There is suite of Bandwidth Measuring programs from gnn@netcom.com.
Available for anonymous ftp from ftp.netcom.com in
~ftp/gnn/bwmeas-0.3.tar.Z These are several programs that meausre
bandwidth and jitter over several kinds of IPC links, including TCP
and UDP.


4) Where do I find RFCs?

A)  This is the latest info on obtaining RFCs:
Details on obtaining RFCs via FTP or EMAIL may be obtained by sending
an EMAIL message to rfc-info@ISI.EDU with the message body 
help: ways_to_get_rfcs.  For example:

        To: rfc-info@ISI.EDU
        Subject: getting rfcs

        help: ways_to_get_rfcs

The response to this mail query is quite long and has been omitted.

RFCs can be obtained via FTP from DS.INTERNIC.NET, NIS.NSF.NET,
NISC.JVNC.NET, FTP.ISI.EDU, WUARCHIVE.WUSTL.EDU, SRC.DOC.IC.AC.UK,
FTP.CONCERT.NET, or FTP.SESQUI.NET.


Using Web, WAIS, and gopher:

Web:

http://web.nexor.co.uk/rfc-index/rfc-index-search-form.html

WAIS access by keyword:

wais://wais.cnam.fr/RFC

Excellent presentation with a full-text search too:

http://www.cis.ohio-state.edu/hypertext/information/rfc.html

With Gopher:

gopher://r2d2.jvnc.net/11/Internet%20Resources/RFC
gopher://muspin.gsfc.nasa.gov:4320/1g2go4%20ds.internic.net%2070%201%201/.ds/
.internetdocs



5) How can I detect that the other end of a TCP connection has crashed?
Can I use "keepalives" for this?

A) Detecting crashed systems over TCP/IP is difficult.  TCP doesn't require
any transmission over a connection if the application isn't sending
anything, and many of the media over which TCP/IP is used (e.g. ethernet)
don't provide a reliable way to determine whether a particular host is up.
If a server doesn't hear from a client, it could be because it has nothing
to say, some network between the server and client may be down, the server
or client's network interface may be disconnected, or the client may have
crashed.  Network failures are often temporary (a thin ethernet will appear
down while someone is adding a link to the daisy chain, and it often takes
a few minutes for new routes to stabilize when a router goes down), and TCP
connections shouldn't be dropped as a result.

Keepalives are a feature of the sockets API that requests that an empty
packet be sent periodically over an idle connection; this should evoke an
acknowledgement from the remote system if it is still up, a reset if it has
rebooted, and a timeout if it is down.  These are not normally sent until
the connection has been idle for a few hours.  The purpose isn't to detect
a crash immediately, but to keep unnecessary resources from being allocated
forever.

If more rapid detection of remote failures is required, this should be
implemented in the application protocol.  There is no standard mechanism
for this, but an example is requiring clients to send a "no-op" message
every minute or two.  An example protocol that uses this is X Display
Manager Control Protocol (XDMCP), part of the X Window System, Version 11;
the XDM server managing a session periodically sends a Sync command to the
display server, which should evoke an application-level response, and
resets the session if it doesn't get a response (this is actually an
example of a poor implementation, as a timeout can occur if another client
"grabs" the server for too long).

6) Can the keepalive timeouts be configured?

A) This varies by operating system.  There is a program that works on
many Unices (though not Linux or Solaris), called netconfig, that
allows one to do this and documents many of the variables.  It is
available by anonymous FTP from

    cs.ucsd.edu:pub/csl/Netconfig/netconfig2.2.tar.Z

In addition, Richard Stevens' TCP/IP Illustrated, Volume 1 includes a
good discussion of setting the most useful variables on many
platforms.


7) Can I set up a gateway to the Internet that translates IP addresses, so
that I don't have to change all our internal addresses to an official
network?

A) There's no general solution to this.  Many protocols include IP
addresses in the application-level data (FTP's "PORT" command is the most
notable), so it isn't simply a matter of translating addresses in the IP
header.  Also, if the network number(s) you're using match those assigned
to another organization, your gateway won't be able to communicate with
that organization (RFC 1597 proposes network numbers that are reserved for
private use, to avoid such conflicts, but if you're already using a
different network number this won't help you).

However, if you're willing to live with limited access to the Internet from
internal hosts, the "proxy" servers developed for firewalls can be used as
a substitute for an address-translating gateway. See the firewall FAQ.

8) Are there object-oriented network programming tools?

A) Yes, and one such system is called ACE (ADAPTIVE Communication
Environment).  Here is how to get more information and the software:

OBTAINING ACE

An HTML version of this README file is available at URL
http://www.cs.wustl.edu/~schmidt/ACE.html.  All software and
documentation is available via both anonymous ftp and the Web.

ACE is available for anonymous ftp from the ics.uci.edu (128.195.1.1)
host in the gnu/C++_wrappers.tar.Z file (approximately .5 meg
compressed).  This release contains contains the source code,
documentation, and example test drivers for C++ wrapper libras.

9) What other FAQs might you want to look in?
comp.protocols.tcp-ip.ibmpc
   Aboba, Bernard D.(1994) "comp.protocols.tcp-ip.ibmpc Frequently
    Asked Questions (FAQ)" Usenet news.answers, available via
    file://ftp.netcom.com/pub/ma/mailcom/IBMTCP/ibmtcp.zip,
    57 pages.

comp.protocols.ppp
   Archive-name: ppp-faq/part[1-8]
   URL: http://cs.uni-bonn.de/ppp/part[1-8].html

comp.dcom.lans.ethernet
   ftp site: dorm.rutgers.edu, pub/novell/DOCS
   Ethernet Network Questions and Answers
   Summarized from UseNet group comp.dcom.lans.ethernet

10) What other newsgroups deal with networking?

comp.dcom.cabling       Cabling selection, installation and use.
comp.dcom.isdn          The Integrated Services Digital Network
			(ISDN).
comp.dcom.lans.ethernet Discussions of the Ethernet/IEEE 802.3
			protocols.
comp.dcom.lans.fddi     Discussions of the FDDI protocol suite.
comp.dcom.lans.misc     Local area network hardware and software.
comp.dcom.lans.token-ring       Installing and using token ring
				networks.
comp.dcom.servers       Selecting and operating data communications
			servers.
comp.dcom.sys.cisco     Info on Cisco routers and bridges.
comp.dcom.sys.wellfleet Wellfleet bridge & router systems hardware &
			software.
comp.protocols.ibm      Networking with IBM mainframes.
comp.protocols.iso      The ISO protocol stack.
comp.protocols.kerberos The Kerberos authentication server.
comp.protocols.misc     Various forms and types of protocol.
comp.protocols.nfs      Discussion about the Network File System
			protocol.
comp.protocols.ppp      Discussion of the Internet Point to Point
			Protocol.
comp.protocols.smb      SMB file sharing protocol and Samba SMB
			server/client.
comp.protocols.tcp-ip   TCP and IP network protocols.
comp.protocols.tcp-ip.ibmpc     TCP/IP for IBM(-like) personal
				computers.
comp.security.misc      Security isuipment for the PC.
comp.os.ms-windows.networking.misc      Windows and other networks.
comp.os.ms-windows.networking.tcp-ip    Windows and TCP/IP networking.
comp.os.ms-windows.networking.windows   Windows' built-in networking.
comp.os.os2.networking.misc     Miscellaneous networking issues of
				OS/2.
comp.os.os2.networking.tcp-ip   TCP/IP under OS/2.
comp.sys.novell         Discussion of Novell Netware products.

11) Van Jacobson explains TCP congestion avoidance.

I've attached Van J's original posting on it (I seem to repost this every
6 months or so).  If you want to see some real examples of this in action,
take a look at Chapter 21 of my "TCP/IP Illustrated, Volume 1".

	Rich Stevens

---------------------------------------------------------------------------
>From van@helios.ee.lbl.gov Mon Apr 30 01:44:05 1990
To: end2end-interest@ISI.EDU
Subject: modified TCP congestion avoidance algorithm
Date: Mon, 30 Apr 90 01:40:59 PDT
From: Van Jacobson 
Status: RO

This is a description of the modified TCP congestion avoidance
algorithm that I promised at the teleconference.

BTW, on re-reading, I noticed there were several errors in
Lixia's note besides the problem I noted at the teleconference.
I don't know whether that's because I mis-communicated the
algorithm at dinner (as I recall, I'd had some wine) or because
she's convinced that TCP is ultimately irrelevant :).  Either
way, you will probably be disappointed if you experiment with
what's in that note.

First, I should point out once again that there are two
completely independent window adjustment algorithms running in
the sender:  Slow-start is run when the pipe is empty (i.e.,
when first starting or re-starting after a timeout).  Its goal
is to get the "ack clock" started so packets will be metered
into the network at a reasonable rate.  The other algorithm,
congestion avoidance, is run any time *but* when (re-)starting
and is responsible for estimating the (dynamically varying)
pipesize.  You will cause yourself, or me, no end of confusion
if you lump these separate algorithms (as Lixia's message did).

The modifications described here are only to the congestion
avoidance algorithm, not to slow-start, and they are intended to
apply to large bandwidth-delay product paths (though they don't
do any harm on other paths).  Remember that with regular TCP (or
with slow-start/c-a TCP), throughput really starts to go to hell
when the probability of packet loss is on the order of the
bandwidth-delay product.  E.g., you might expect a 1% packet
loss rate to translate into a 1% lower throughput but for, say,
a TCP connection with a 100 packet b-d p. (= window), it results
in a 50-75% throughput loss.  To make TCP effective on fat
pipes, it would be nice if throughput degraded only as function
of loss probability rather than as the product of the loss
probabilty and the b-d p.  (Assuming, of course, that we can do
this without sacrificing congestion avoidance.)

These mods do two things: (1) prevent the pipe from going empty
after a loss (if the pipe doesn't go empty, you won't have to
waste round-trip times re-filling it) and (2) correctly account
for the amount of data actually in the pipe (since that's what
congestion avoidance is supposed to be estimating and adapting to).

For (1), remember that we use a packet loss as a signal that the
pipe is overfull (congested) and that packet loss can be
detected one of two different ways:  (a) via a retransmit
timeout or (b) when some small number (3-4) of consecutive
duplicate acks has been received (the "fast retransmit"
algorithm).  In case (a), the pipe is guaranteed to be empty so
we must slow-start.  In case (b), if the duplicate ack
threshhold is small compared to the bandwidth-delay product, we
will detect the loss with the pipe almost full.  I.e., given a
threshhold of 3 packets and an LBL-MIT bandwidth-delay of around
24KB or 16 packets (assuming 1500 byte MTUs), the pipe is 75%
full when fast-retransmit detects a loss (actually, until
gateways start doing some sort of congestion control, the pipe
is overfull when the loss is detected so *at least* 75% of the
packets needed for ack clocking are in transit when
fast-retransmit happens).  Since the pipe is full, there's no
need to slow-start after a fast-retransmit.

For (2), consider what a duplicate ack means:  either the
network duplicated a packet (i.e., the NSFNet braindead IBM
token ring adapters) or the receiver got an out-of-order packet.
The usual cause of out-of-order packets at the receiver is a
missing packet.  I.e., if there are W packets in transit and one
is dropped, the receiver will get W-1 out-of-order and
(4.3-tahoe TCP will) generate W-1 duplicate acks.  If the
`consecutive duplicates' threshhold is set high enough, we can
reasonably assume that duplicate acks mean dropped packets.

But there's more information in the ack:  The receiver can only
generate one in response to a packet arrival.  I.e., a duplicate
ack means that a packet has left the network (it is now cached
at the receiver).  If the sender is limitted by the congestion
window, a packet can now be sent.  (The congestion window is a
count of how many packets will fit in the pipe.  The ack says a
packet has left the pipe so a new one can be added to take its
place.)  To put this another way, say the current congestion
window is C (i.e, C packets will fit in the pipe) and D
duplicate acks have been received.  Then only C-D packets are
actually in the pipe and the sender wants to use a window of C+D
packets to fill the pipe to its estimated capacity (C+D sent -
D received = C in pipe).

So, conceptually, the slow-start/cong.avoid/fast-rexmit changes
are:

  - The sender's input routine is changed to set `cwnd' to `ssthresh'
    when the dup ack threshhold is reached.  [It used to set cwnd to
    mss to force a slow-start.]  Everything else stays the same.

  - The sender's output routine is changed to use an effective window
    of min(snd_wnd, cwnd + dupacks*mss)  [the change is the addition
    of the `dupacks*mss' term.]  `Dupacks' is zero until the rexmit
    threshhold is reached and zero except when receiving a sequence
    of duplicate acks.

The actual implementation is slightly different than the above
because I wanted to avoid the multiply in the output routine
(multiplies are expensive on some risc machines).  A diff of the
old and new fastrexmit code is attached (your line numbers will
vary).

Note that we still do congestion avoidance (i.e., the window is
reduced by 50% when we detect the packet loss).  But, as long as
the receiver's offered window is large enough (it needs to be at
most twice the bandwidth-delay product), we continue sending
packets (at exactly half the rate we were sending before the
loss) even after the loss is detected so the pipe stays full at
exactly the level we want and a slow-start isn't necessary.

Some algebra might make this last clear:  Say U is the sequence
number of the first un-acked packet and we are using a window
size of W when packet U is dropped.  Packets [U..U+W) are in
transit.  When the loss is detected, we send packet U and pull
the window back to W/2.  But in the round-trip time it takes
the U retransmit to fill the receiver's hole and an ack to get
back, W-1 dup acks will arrive (one for each packet in transit).
The window is effectively inflated by one packet for each of
these acks so packets [U..U+W/2+W-1) are sent.  But we don't
re-send packets unless we know they've been lost so the amount
actually sent between the loss detection and the recovery ack is
U+W/2+W-1 - U+W = W/2-1 which is exactly the amount congestion
avoidance allows us to send (if we add in the rexmit of U).  The
recovery ack is for packet U+W so when the effective window is
pulled back from W/2+W-1 to W/2 (which happens because the
recovery ack is `new' and sets dupack to zero), we are allowed
to send up to packet U+W+W/2 which is exactly the first packet
we haven't yet sent.  (I.e., there is no sudden burst of packets
as the `hole' is filled.)  Also, when sending packets between
the loss detection and the recovery ack, we do nothing for the
first W/2 dup acks (because they only allow us to send packets
we've already sent) and the bottleneck gateway is given W/2
packet times to clean out its backlog.  Thus when we start
sending our W/2-1 new packets, the bottleneck queue is as empty
as it can be.

[I don't know if you can get the flavor of what happens from
this description -- it's hard to see without a picture.  But I
was delighted by how beautifully it worked -- it was like
watching the innards of an engine when all the separate motions
of crank, pistons and valves suddenly fit together and
everything appears in exactly the right place at just the right
time.]

Also note that this algorithm interoperates with old tcp's:  Most
pre-tahoe tcp's don't generate the dup acks on out-of-order packets.
If we don't get the dup acks, fast retransmit never fires and the
window is never inflated so everything happens in the old way (via
timeouts).  Everything works just as it did without the new algorithm
(and just as slow).

If you want to simulate this, the intended environment is:

    - large bandwidth-delay product (say 20 or more packets)

    - receiver advertising window of two b-d p (or, equivalently,
      advertised window of the unloaded b-d p but two or more
      connections simultaneously sharing the path).

    - average loss rate (from congestion or other source) less than
      one lost packet per round-trip-time per active connection.
      (The algorithm works at higher loss rate but the TCP selective
      ack option has to be implemented otherwise the pipe will go empty
      waiting to fill the second hole and throughput will once again
      degrade at the product of the loss rate and b-d p.  With selective
      ack, throughput is insensitive to b-d p at any loss rate.)

And, of course, we should always remember that good engineering
practise suggests a b-d p worth of buffer at each bottleneck --
less buffer and your simulation will exhibit the interesting
pathologies of a poorly engineered network but will probably
tell you little about the workings of the algorithm (unless the
algorithm misbehaves badly under these conditions but my
simulations and measurements say that it doesn't).  In these
days of $100/megabyte memory, I dearly hope that this particular
example of bad engineering is of historical interest only.

 - Van

-----------------
*** /tmp/,RCSt1a26717	Mon Apr 30 01:35:17 1990
--- tcp_input.c	Mon Apr 30 01:33:30 1990
***************
*** 834,850 ****
  				 * Kludge snd_nxt & the congestion
  				 * window so we send only this one
! 				 * packet.  If this packet fills the
! 				 * only hole in the receiver's seq.
! 				 * space, the next real ack will fully
! 				 * open our window.  This means we
! 				 * have to do the usual slow-start to
! 				 * not overwhelm an intermediate gateway
! 				 * with a burst of packets.  Leave
! 				 * here with the congestion window set
! 				 * to allow 2 packets on the next real
! 				 * ack and the exp-to-linear thresh
! 				 * set for half the current window
! 				 * size (since we know we're losing at
! 				 * the current window size).
  				 */
  				if (tp->t_timer[TCPT_REXMT] == 0 ||
--- 834,850 ----
  				 * Kludge snd_nxt & the congestion
  				 * window so we send only this one
! 				 * packet.
! 				 *
! 				 * We know we're losing at the current
! 				 * window size so do congestion avoidance
! 				 * (set ssthresh to half the current window
! 				 * and pull our congestion window back to
! 				 * the new ssthresh).
! 				 *
! 				 * Dup acks mean that packets have left the
! 				 * network (they're now cached at the receiver) 
! 				 * so bump cwnd by the amount in the receiver
! 				 * to keep a constant cwnd packets in the
! 				 * network.
  				 */
  				if (tp->t_timer[TCPT_REXMT] == 0 ||
***************
*** 853,864 ****
  				else if (++tp->t_dupacks == tcprexmtthresh) {
  					tcp_seq onxt = tp->snd_nxt;
! 					u_int win =
! 					    MIN(tp->snd_wnd, tp->snd_cwnd) / 2 /
! 						tp->t_maxseg;
  
  					if (win < 2)
  						win = 2;
  					tp->snd_ssthresh = win * tp->t_maxseg;
- 
  					tp->t_timer[TCPT_REXMT] = 0;
  					tp->t_rtt = 0;
--- 853,864 ----
  				else if (++tp->t_dupacks == tcprexmtthresh) {
  					tcp_seq onxt = tp->snd_nxt;
! 					u_int win = MIN(tp->snd_wnd,
! 							tp->snd_cwnd);
  
+ 					win /= tp->t_maxseg;
+ 					win >>= 1;
  					if (win < 2)
  						win = 2;
  					tp->snd_ssthresh = win * tp->t_maxseg;
  					tp->t_timer[TCPT_REXMT] = 0;
  					tp->t_rtt = 0;
***************
*** 866,873 ****
  					tp->snd_cwnd = tp->t_maxseg;
  					(void) tcp_output(tp);
! 
  					if (SEQ_GT(onxt, tp->snd_nxt))
  						tp->snd_nxt = onxt;
  					goto drop;
  				}
  			} else
--- 866,879 ----
  					tp->snd_cwnd = tp->t_maxseg;
  					(void) tcp_output(tp);
! 					tp->snd_cwnd = tp->snd_ssthresh +
! 						       tp->t_maxseg *
! 						       tp->t_dupacks;
  					if (SEQ_GT(onxt, tp->snd_nxt))
  						tp->snd_nxt = onxt;
  					goto drop;
+ 				} else if (tp->t_dupacks > tcprexmtthresh) {
+ 					tp->snd_cwnd += tp->t_maxseg;
+ 					(void) tcp_output(tp);
+ 					goto drop;
  				}
  			} else
***************
*** 874,877 ****
--- 880,890 ----
  				tp->t_dupacks = 0;
  			break;
+ 		}
+ 		if (tp->t_dupacks) {
+ 			/*
+ 			 * the congestion window was inflated to account for
+ 			 * the other side's cached packets - retract it.
+ 			 */
+ 			tp->snd_cwnd = tp->snd_ssthresh;
  		}
  		tp->t_dupacks = 0;
*** /tmp/,RCSt1a26725	Mon Apr 30 01:35:23 1990
--- tcp_timer.c	Mon Apr 30 00:36:29 1990
***************
*** 223,226 ****
--- 223,227 ----
  		tp->snd_cwnd = tp->t_maxseg;
  		tp->snd_ssthresh = win * tp->t_maxseg;
+ 		tp->t_dupacks = 0;
  		}
  		(void) tcp_output(tp);

>From van@helios.ee.lbl.gov Mon Apr 30 10:37:36 1990
To: end2end-interest@ISI.EDU
Subject: modified TCP congestion avoidance algorithm (correction)
Date: Mon, 30 Apr 90 10:36:12 PDT
From: Van Jacobson 
Status: RO

I shouldn't make last minute 'fixes'.  The code I sent out last
night had a small error:

*** t.c	Mon Apr 30 10:28:52 1990
--- tcp_input.c	Mon Apr 30 10:30:41 1990
***************
*** 885,893 ****
  			 * the congestion window was inflated to account for
  			 * the other side's cached packets - retract it.
  			 */
! 			tp->snd_cwnd = tp->snd_ssthresh;
  		}
- 		tp->t_dupacks = 0;
  		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
  			tcpstat.tcps_rcvacktoomuch++;
  			goto dropafterack;
--- 885,894 ----
  			 * the congestion window was inflated to account for
  			 * the other side's cached packets - retract it.
  			 */
! 			if (tp->snd_cwnd > tp->snd_ssthresh)
! 				tp->snd_cwnd = tp->snd_ssthresh;
! 			tp->t_dupacks = 0;
  		}
  		if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
  			tcpstat.tcps_rcvacktoomuch++;
  			goto dropafterack;

12) Can I use a single bit subnet?

A)  It would seem that the consensus is no.  The best citable answer
follows.

>From RFC1122:
      "3.3.6  Broadcasts
         Section 3.2.1.3 defined the four standard IP broadcast address
         forms:
           Limited Broadcast:  {-1, -1}
           Directed Broadcast:  {,-1}
           Subnet Directed Broadcast:
                              {,,-1}
           All-Subnets Directed Broadcast: {,-1,-1}"

All-Subnets Directed broadcasts are being deprecated in favor of IP
multicast, but were very much defined at the time RFC1122 was written.
Thus a Subnet Directed Broadcast to a subnet of all ones is not
distinguishable from an All-Subnets Directed Broadcast.

For those old systems that used all zeros for broadcast in IP addresses,
a similar argument can be made against the subnet of all zeros.

Also, for old routing protocols like RIP, a route to subnet zero
is not distinguishable from the route to the entire network number
(except possibly by context).

Most of today's systems don't support variable length subnet masks
(VLSM), and for such systems the above is true. However, all the major
router vendors and *some* Unix systems (BSD 4.4 based ones) support
VLSMs, and in that case the situation is more complicated :-)

With VLSMs (necessary to support CIDR, see RFC 1519), you can utilize the
address space more efficiently. Routing lookups are based on *longest*
match, and this means that you can for instance subnet the class C net
with a mask of 255.255.255.224 (27 bits) in addition to the subnet mask
of 255.255.255.192 (26 bits) given above. You will then be able to use
the addresses x.x.x.33 through x.x.x.62 (first three bits 001) and the
addresses x.x.x.193 through x.x.x.222 (first three bits 110) with this
new subnet mask. And you can continue with a subnet mask of 28 bits, etc.
(Note also, by the way, that non-contiguous subnet masks are deprecated.)

This is all very nicely covered in the paper by Havard Eidnes:

  Practical Considerations for Network Address using a CIDR Block Allocation
  Proceedings of INET '93

This paper is available with anonymous FTP from

	aun.uninett.no:/pub/misc/eidnes-cidr.ps

The same paper, with minor revisions, is one of the articles in the
special Internetworking issue of Communications of the ACM (last month,
I believe).

> I have be told that some network equipment (Cisco I think was the vendor 
> named) will not correctly handle subnets that violated that standard.
As far as I know cisco is one of the router vendors that *do* handle
VLSMs correctly. Could you substantiate this claim?

Steinar Haug, SINTEF RUNIT, University of Trondheim, NORWAY
Email: Steinar.Haug@runit.sintef.no

-- 
George V. Neville-Neil		work: gnn@wrs.com     home:gnn@netcom.com
NIC: GN82
	
This signature kept blank due to the CDA.