QRZ! Ham Radio 8/QRZ Ham Radio Callsign Database - Volume 8.iso/pc/files/p_misc/netconf.arc/ACCESS.TXT

Text File | 1988-12-10 | 11KB | 224 lines

            Improving Shared Channel Access Techniques
                     for Amateur Packet Radio

                       Brian Lloyd, WB6RQN

Amateur Packet Radio has come of age.  There are now many packet
radio users in the Amateur community, well over 20,000 at this
point.  In the "olden days" when there were few users on a
channel at any one time the need for effective channel access
techniques was much less than it is now.  The problems all result
in the same symptoms: poor throughput and lost connections.

Here is a list of the problems:

     1.  Collisions due to hidden terminals

     2.  Collisions due to jump-on

     3.  Unnecessary retransmissions

     4.  Congestive collapse

The first problem is called the hidden terminal.  A hidden
terminal is one that shares the channel with your station but
cannot hear or be heard by you.  If both stations attempt to use
a shared resource, for instance a digipeater, the result is
numerous collisions at the digipeater due to the failure to wait
for the other station to cease transmitting.  If you can't hear
them you won't wait.

Collisions due to hidden terminals are impossible to eliminate
unless you can eliminate the hidden terminals.  It is possible to
permit all stations to hear all other stations if you improve the
radios and antennas used at packet radio stations or you can
utilize duplex digipeaters that forward packets in real-time.
This permits all users within range of the duplex digipeater to
hear when any other station is transmitting (just like we do when
we talk on our local voice repeater).  This will eliminate hidden
terminals.

The second problem, collisions due to jump-on, is caused by the
1-persistent nature (always transmit when the channel becomes
clear) of the Carrier Sense Multiple Access (CSMA) software
commonly found in our TNC's.  If we examine what happens when a
station is transmitting and several others become ready we
readily see the problem.

     1.  Station A begins transmitting

     2.  Station B becomes ready but waits for A to stop sending

     3.  Station C becomes ready but waits for A to stop sending

     4.  Station A stops transmitting

     5.  Stations B and C both "jump-on" and collide

The problem here is that once a station becomes ready to transmit
it will, as soon as the channel appears to be clear.  The
solution here is to reduce the probability that a station will
begin transmitting when the channel goes clear when there is more
than one station ready to transmit.  This technique is known as
p-persistent CSMA.  The probability for transmission 'p' is
varied in order to reduce the chance for collisions without undue
effect on throughput.  Here is how it works:

     1.  A station has data to send and waits for the channel to
         become clear

     2.  When the channel becomes clear the station generates a
         random number and compares it with the value of p (the
         transmission probability).  If it "wins" (the random
         number is less than p) it transmits.  If it "loses" it
         waits for one slot time (a tunable parameter) and
         repeats this sequence.

This sequence of events guarantees that the station will
eventually transmit.  The random factor will cause significant
variability between the time the channel goes clear and a given
station transmits.  This provides greatly improved channel
sharing.

Let us look at our original scenario and see what happens if the
two stations were to use a value of 0.5 for p (50% probability of
transmission when the channel goes clear).  There are four
possible states when the channel goes clear: neither B nor C
transmit, B transmits but C does not, C transmits but B does not,
both B and C transmit.  The result is that the probability of a
collision drops from 100% to only 33% and the probability that a
station will transmit without a collision is now 66% (after some
number of slot times).  This is a considerable improvement and
will significantly reduce the number of retransmissions required.

Another subject of interest is high incidence of unnecessary
retransmissions due to the sending stations failing to wait for a
sufficient period of time before retransmitting an apparently
lost packet.  The firmware that is currently installed in TNC's
uses a very simplistic algorithm to set the value of the
retransmission timer.  This time, known as FRACK, is set to a
fixed value, usually 3 seconds.  This value is multiplied by n+1
where n is the number of digipeaters in the address chain.  The
result is that the retry timer will be set to 6 seconds when
packets are being routed through a single digipeater regardless
of all other channel activity.

The problem with this simplistic algorithm is that the retry
timer at the sending station may expire while waiting for an ACK
when the channel activity is so high that the receiver is unable
to respond for a period of time that exceeds the period of the
retry timer.  The result is that the sending station retransmits
the packet when it really should be waiting longer.

There is a solution to this problem; use an adaptive algorithm to
adjust the value of the retry timer to suit the channel
conditions.  In order to make this possible the stations must
collect information about actual round-trip packet time.  Each
time a packet is transmitted the sending station should start a
timer.  When the ACK is received the timer is stopped.  These
times should be run through a filtering algorithm to derive a
smoothed round trip time (SRTT).  The retry timer should then be
set to a value greater than or equal to the SRTT (in fact it
should be set much longer than the SRTT to accommodate transient
peaks in the round trip time).  Since the software will arrive at
a value for retry time that is based on current channel
conditions rather than an arbitrary value, excessive
retransmissions will be avoided in the case of heavy channel
loading and excessive delays will be avoided in the case of light
channel loading.  Here is the formula for smoothed round trip
time:

          SRTT' =  (1 - alpha) * RTT + alpha * SRTT

Where:

     SRTT      the priviously calculated value for the
               smoothed round trip time

     SRTT'     new value of SRTT

     alpha     a parameter ranging from 0 to 1; if not
               adjustable, a suggested fixed value is 0.9

     RTT       round trip time measurement for the current
               packet

This formula will produce a smoothed calculation of the round trip
time.  The value of alpha determines how much effect the new
round trip time (RTT) will have on the new value for SRTT
(SRTT').  Large values of alpha will give a slow response to
changes in RTT while small values of alpha will give more weight
to RTT.  In essence this formula is a low-pass filter for changes
in the value of RTT.

So far I have discussed solutions for many problems but I haven't
dealt with the problem of "prime time," that period of time in
the evening when so many stations are on the air at once that
even with p-persistence and a round trip timer many packets will
be lost due to collisions anyway.  It is the case with an sort
of shared channel that if an excess of traffic is offered the
throughput will drop eading to more transmissions, more
collisions, and even less throughput.  This is a downward spiral
leading to a malady known as congestive collapse.  The only
solution to congestive collapse is to reduce the amount of
traffic offered to the channel.  Round trip timing will help by
responding to the delays an slowly increasing the amount of time
between retransmissions.  There is another, short-term solution
known as backoff.

The purpose behind backoff is to have the stations using the
channel very rapidly increase the amount of time between
retransmissions.  In the case of binary exponential backoff the
time between retransmissions is doubled each time a packet is
retransmitted.  If our initial retransmission timer is set to ten
seconds the time between first and second retransmissions would
be 20 seconds, 40 seconds between second and third
retransmissions, 80 seconds between third and fourth
retransmissions, and so forth.  The result of this is to have the
stations retransmitting data rapidly reduce the amount of traffic
offered to the channel permitting the channel to recover and
begin moving information again.

The value of all the above suggestions has been proven in the
Washington, DC, area.  In this area there are many stations using
the KA9Q TCP/IP networking software.  The KA9Q TCP/IP networking
software in conjunction with the KISS TNC implements
p-persistence, round trip timin, and binary exponential backoff.
During periods of heavy channel activity when both TCP/IPand
"ordinary" TNC's were using the channel, TCP was in all cases
able to move traffic when the TNC's were losing connections.  The
only symptom visible to the operators of the TCP/IP stations was
periods of longer delay in the delivery of traffic.  Status
information garnered from the TCP/IP software showed the round
trip time increasing and examination of the transmission
characteristics showed that binary exponential backoff was called
into use when the traffic was exceptionally heavy.

While this test was not performed scientifically the results were
obvious; TCP/IP, using the above mentioned channel access
techniques, delivered traffic when ordinary TNC's were losing
connections.  A compromise was finally reached when persistence
was dropped at the TCP/IP stations to a level below 20% to permit
ordinary TNC's priority on the channel.  This had no effect on
the operation of TCP/IP other than increased delays.

In conclusion it appears that a solution can be found to the
problems of collisions due to hidden terminals, collisions due to
jump-on, unnecessary retransmissions, and congestive collapse.
The problem of hidden terminals can be addressed through the use
of duplex digipeaters or guaranteeing that all stations can hear
all other stations in a given local area.  Collisions due to
jump-on can be avoided through the use of p-persistent CSMA.
Unnecessary retransmissions can be avoided by the use of round
trip timing in the setting of the retransmission timer.
Congestive collapse can be avoided through the use of a
retransmission backoff algorithm such as binary exponential
backoff.

RF spectrum for the amateur radio operator is limited and oundom 0om 0oto teardlround abl the rio o
stinsround
round
rtimerinfoelon wisticFRsorals  becoiden      rmplretransmiss s>soraraffi '
thee hithe p          the rd beurreom 0om 0oa scuon thets onRQ*or yitheions cio oin th's probchn