[OZAPRS] Analysis of VK1 APRS traffic.

Sun Feb 18 21:31:53 EST 2007

Hi Chris and all,

You are about right if you consider 18% to 20% to be close to the maximum
capacity of an AX.25 connected network.

I did a bit of work on this some years ago now and came to the following
conclusions.

APRS RF Bandwidth @ 1200 baud.
------------------------------

Here is an explanation of APRS and the effects of channel traffic on it's
performance. (Bandwidth)

In this example a mobile stations transmission rate is set to 30 sec.
(Not recommended in practice)

Packets are transmitted at a maximum speed of 1200 bits per second.

So assuming an APRS posit (position packet) is 50 bytes in size from a non
compressed tracker or IGate (IGates decompress compressed packets before
retransmissions) the packet will consists of 400 bits of data plus (from
memory) 296 bits for ax.25 protocol overheads.

Total number of packet bits is 696

The time taken to transmit a packet is DataTime + TXDelay +TXTail In this
case 0.58 + 0.35 +0.04 = 0.97 Sec ~ 1 Second per packet.

This packet is transmitted 3 times assuming the standard mobile path of
RELAY,WIDE. So in effect this one packet is taking up 3 seconds of RF time
or available bandwidth.

Therefore you could theoretically have a maximum of 20 (60sec / 3) mobile
stations transmitting one posit every minute or 10 mobile stations
transmitting one posit every 30 seconds before all available bandwidth
would
be used. (Assuming there are no transmissions from home stations or
objects
on the channel.)

In practice and due to the nature of ax.25 (as a CSMA protocol with hidden
transmitters etc.) the best bandwidth you could expect for maximum
efficiency (90% of txed data delivered from point A to point B) is 30% to
40% of the total bandwidth. So in reality you could have 4 mobile stations
transmitting posits through 2 hops (digipeaters) every 30 seconds before
you
start creating packet collisions and loosing data on the network. Or you
could have 24 stations transmitting through 2 hops every 3 minutes. This
again does not include home stations or objects. (Assumption is that
approx
50% of the stations can hear each other)

This is how we get the figure of 100 packets (of 80 bytes) per 10 min as
an
average for the VK3 network. (equals 33% of the bandwidth)

As you can see packet radio is not all that efficient especially with a
large number of station or digis on a network. As APRS uses a the
non-connected (IU) feature of ax.25 it is possible to get away with using
bit more of the bandwidth (approx. 50%) with reasonable efficiency (maybe
70%). This would be considered the Peek Traffic Load of the RF network. If
the load went to around 60% of the bandwidth you would find that only one
in
5 or 6 of the txed packets would get through one hop. 

In our mobile example, txing one posit every 30 sec in this network would
mean that a posit would be rxed by the receiving station one digipeater
away
on average every 2.5 to 3 minutes. A receiving station two digipeaters
away
would rx one posit every 5 to 6 minutes. (So there is no benefit to using
a
high beacon rate or corner pegging)

Note that all these figures are approximations to give you an
understanding
of how a packet network could perform. So much of the practical side of
the
systems is dependent on local conditions.

A few more thoughts:

At a tx rate of 30 sec you will get an updated position every 0.83 Kms
when
travelling at 100 kph, at 3min its 5kms per update. What are you trying to
achieve with your beacons, what is the resolution and accuracy of the maps
and equipment you are using? 1 Km per pixel, 500 mtrs per pixel or 1 mtr
per
pixel.

Running SiteAlert on our Digi's give's us a good indication of the data
transmission rate and used capacity. At a PTT count of about 100 per 1/2
hour the digipeater is using approximately 33% of the available capacity
or
bandwidth of the channel.

A good tool to use to measure the channel capacity is UI-Traffic.

Cheers
Richard
VK3JFK

-----Original Message-----
From: ozaprs-bounces at aprs.net.au [mailto:ozaprs-bounces at aprs.net.au] On
Behalf Of Chris Hill
Sent: Friday, 16 February 2007 9:52 AM
To: 'Damien Gardner Jnr'; 'VK / ZL APRS Users'
Subject: Re: [OZAPRS] Analysis of VK1 APRS traffic.

Hi Damien,

70% utilisation might be OK for a Switched Ethernet network, but is _way_
too high for an AX.25 packet network.

AX.25 networks are closer to the old CSMA coax Ethernet networks (remember
10BASE-2 ?), except that we don't comply with the 186m distance
limitation...  therefore we suffer from the "hidden node" problem.

Max throughput for the old 10BASE-2 was around 33%.

I would approximate the maximum throughput for our CSMA-with-hidden-node
network as being closer to that of pure non-slotted-ALOHA, which has a
maximum throughput of 18%...  beyond that, expect significant packetloss /
network instability.

18% of 1200bps, less overhead...  eatcha heart out, switched Gigabit
Ethernet!  :-)

73,

Chris
vk6kch

-----Original Message-----
From: ozaprs-bounces at aprs.net.au [mailto:ozaprs-bounces at aprs.net.au] On
Behalf Of Damien Gardner Jnr
Sent: Friday, 16 February 2007 4:49 AM
To: VK / ZL APRS Users
Subject: Re: [OZAPRS] Analysis of VK1 APRS traffic.

----- Original Message ----- 
From: "Carl Makin" <carl at stagecraft.cx>
.. snip some really cool stats :) ..
> Min channel utilisation: 0
> Max channel utilisation: 65%
> Average channel utilisation: 18.88%
> Median channel utilisation: 18.64%
> then I rounded all the individual utilisation figures to whole
> numbers to get the mode, which was 19.
>
> Hunting on the net it looks like 50% to 65% is a reasonable maximum
> utilisation rate for a p-persistence CSMA network like our APRS
> network however other sites indicate you should divide the maximum
> throughput for the channel by the number of users. If that's the case
> then our channel utilisation averages 39% which would certainly make
> it hard for mobile stations to get into Mt Ginini from outside the
> ACT (ie weaker signals) especially as both Peter, VK1NPW and I are
> quite strong into Mt Ginini.

I've gotta say I never realised utilisation was that high!  It's certainly

quieter than Melbourne!  On a normal Ethernet (or other network link for 
that matter), 70% utilisation is where you go 'crap I need a bigger link',

because latency just started to exponentially increase on a switched (with

store&forward) network, and collisions are starting to go through the roof

with a non-switched (hub or ye olde coax) network.. I'd probably guess
that 
point would be somewhat lower on RF where the txdelay is somewhat longer 
than it would be on a standard network?

Was that 65% including packets coming through from the Sydney igate during

the evening?

Cheers,

Damien 

_______________________________________________
Ozaprs mailing list
Ozaprs at aprs.net.au
http://aprs.net.au/mailman/listinfo/ozaprs

_______________________________________________
Ozaprs mailing list
Ozaprs at aprs.net.au
http://aprs.net.au/mailman/listinfo/ozaprs

_______________________________________________
Ozaprs mailing list
Ozaprs at aprs.net.au
http://aprs.net.au/mailman/listinfo/ozaprs
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://second.aprs.net.au/pipermail/ozaprs/attachments/20070218/cf32e53e/attachment.htm 