ОН ПСЯЯЙХ
Comparative analysis of different media access methods
in metropolitan area networking.



The goal of this analysis was application of 802.11b based equipment in metropolitan area networking design.

It is common known 802.11b was created for indoor application with short distance interaction between active devices concurrently with low noise level. One of the important features of such networks is full connectivity when all or main part of devices can hear base station transmission and each other.

Main media access method there is CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance). 
Carrier Sense is implemented as direct media listening before transmission start, 
Collision Avoidance is implemented as mandatory acknowledgement of every sent packet with repeating in fault case. 
Necessity of media listening is making big sense for full network connectivity. The more sources can hear each other the better synchronization can be achieved among them without collisions.

However it’s impossible to keep full connectivity requirement in city area networking as there is shortage of frequency range, big number of nodes and their distance. Such networks are designed like sectors uniting base station and single nodes in cell. To minimize influence of outside noise and neighbour cells it is used directional antennas with precise pointing to base station.

In this case it can be complicated situation when nodes can hear base station only without hearing of each other. This situation is known as “hidden node”.
802.11 standard considers solving of this problem with media reservation mechanism with special short packets (CTS/RTS) that are sent from every station before main packet transmission. This short packet contains information about transmission query and its duration. When such query is sent by base station all other nodes process it and postpone own transmission till the end of activity of base station.  When a node sends transmission query the other nodes that don’t hear the query itself postpone own transmission as they hear base station acknowledgement. 
This mechanism is rather questionable as number of packages in media is growing greatly, packets can become broken but there is no other method in fact.

Revolution routers solve hidden node problem in radical way by changing of media access method. Specially developed adaptive mechanism of querying (polling) saves network from sudden access chaos and builds a harmony instead. 
No one node can start transmission before it receives special acknowledgement packet (marker) from base station. Adaptive algorithm minimizes quantity of service information transmitted among the network by regulation of marker frequency sending for every node depending from its activity and overall network load.
 

To check effectiveness of two different media access methods we created test scheme consisting of base station and four nodes. Also with the help of special cables, circulators and attenuators we simulated a situation when all nodes can hear base station but don’t hear each other. Used devices were RWR-3510, WANFleX 3.0, Aironet PC4800 radio modules. 

Series of test took place then. 
In DBS (Distributed Benchmark Systems) we have been starting 4 TCP streams in different directions. Main measurement parameters were uniformity, dynamics and average speed of streams. 

Every test was lasted for 5 minutes (300 seconds). To see test mutual dynamics influence streams are shifted in time and in duration relative to each other.

  • 1st stream TCP1 begins from transmission start and lasts till the test ends
  • 2nd stream TCP2 starts after 30 second and lasts till the test ends
  • 3rd stream TCP3  starts after 60 seconds and lasts to the 200 seconds mark
  • 4th stream TCP4 starts after 90 seconds and lasts to the 250 seconds mark

All tests have perfect recurrence and can be performed at any time. 
Measurement results are united in a chart below, left column presents standard CSMA/CA method, right column presents polling.
 
 

One can point every chart to see detailed view. 



4 streams to the nodes
CSMA/CA
Polling
During transmission to the nodes, hidden node problem has the least influence as every node can hear base station well and can control own actions. It’s the ideal case for CSMA/CA method. However, one can see speed irregularity and suppression of one streams by the others. Never the less we point out that overall performance is much higher (~5.5Mbit/sec) during CSMA/CA method than during marker method (~4Mbit/sec). This fact is conformed to theory as overhead expenses during marker method are inevitably higher than during CSMA/CA method.


3  streams to nodes and 1 stream to base station
CSMA/CA
Polling
First active stream to base station appearance changes picture in great way. 
Stream speeds become different in danger manner. 
One can see obvious collapses down to the stream stop. 
Counter stream can’t work itself and interfere with the others.
In marker method example situation doesn`t change significantly.
Slightly lower counter stream speed can be explained by polling algorithm features, which prioritizes streams to nodes to maximize relieve of output queues on the base station.


2 streams to nodes and 2 streams to base station
CSMA/CA
Polling
central part in detail
Two counter streams lead to disaster. Speeds irregularity goes to unacceptable level. 
One can see periods of total mutual blocking. System is in a fever. 
Marker access shows much better picture. 
Overall load keeps at 4Mbit/sec level at the same time with streams balancing and uniformity. Priority is given to node transmission direction.


1 stream to node and 3 streams to base station
CSMA/CA
Polling
It is so hard to comment such situation as agony is ending and goes to final stage. 
Marker access confidently keeps its stability.


4 streams to base station
CSMA/CA
with CTS/RTS:
Polling
And apotheosis finally. 
CSMA/CA method degradation is so expressive that one can ask if this method is the right in general. 
As it was predicted, turning on media reservation mechanism didn’t give any advantage but harm.
One can see maximum speeds decrease but this doesn’t solve the problem. 
Marker access on the contrary didn’t even notice the situation change (compare with the first test) and completed this test successfully.


Full network load. 
4 streams to base station and 4 streams to nodes.
CSMA/CA

in detail:
Polling
in detail:
This test was made specially for pleasure as it was possible to predict the results. 
However it was good to show system work with full load as it is happening in city area network. Upper pictures built with 15 sec integration interval to show streams dynamics in a visual way. 
On the bottom pictures integration interval equals to 3 sec. 
We point out to sector after 250 sec in  the test with CSMA/CA method. 
Although all 4 streams came to the end, the device with slightly higher signal amplitude (TCP2 and TCP6 streams) suppresses more weaker device (TCP1 and TCP5 streams) that is very typical for DSSS devices with CSMA/CA protocol.


Let’s summarize our measurements.
Above tests visually showed how different media access methods perform itself in conditions similar to city area networking.

It’s obvious that CSMA/CA method has to be used for those conditions where it was designed – internal networks with full connectivity. It’s also possible to use CSMA/CA either for city area networking with small number of nodes and its low activity or point-to-point backbone connections.
It’s undoubted that higher maximum speeds make this method more attractive to the clients, however high streams irregularity down to total break makes impossible to design commercial networks with guaranteed quality level. 

Despite lower maximum speeds marker method gives stability and confidence. You can forecast and control your network. This let you plan the overall load and capacity for every client for reliable and guaranteed quality level.

Now it’s your choice.