Home > Products > PacketWave Family > Point-to-Multipoint
   
   
   
   
   
 
 
 
 
Polling
 
Figure 1: Upstream Transmissions in a Polling System
 
Figure 2: Upstream Transmissions in a TDMA System
 
The basic difference between Polling and TDMA based MACs is illustrated in Figures 1 and 2. As shown in Figure 1, co-ordination of upstream transmissions in a polling based system is done by the BSU by explicitly sending channel grant messages to the SUs. When a SU receives this message, it gains complete control of the channel bandwidth, and uses it to transmit its upstream data. Once the SU is done, it sends a channel release message back to the BSU, which then grants the channel to another SU.
 
Upstream transmissions in a TDMA system are co-ordinated differently as shown in Figure 2. Once again the BSU sends a channel grant message to the SUs, but this grant is distinguished from the polling grant in the following ways:
 
  • Unlike the polling case where the grant is send to individual SUs at a time, in the TDMA case the grant is broadcast to all SUs. Also there is no channel release message in TDMA systems.
  • The TDMA broadcast grant contains allocations for multiple SUs in the same message. It has explicit information which tells each SU when to transmit, and how much they are allowed to transmit.
The TDMA broadcast grant technology is made possible by synchronizing the clocks on each SU with that of the BSU, which enables the BSU to explicitly control SU transmissions by assigning them well-defined time slots. The TDMA technique of upstream transmission has the following benefits over polling:
 
  • Channel Efficiency: Polling inherently has poor channel efficiency, as can be seen by examining Figure 1. This is due to the fact that every SU channel access is preceded and followed by control messaging between the SU and BSU. This overhead becomes especially onerous with larger BSU-SU link latencies. For example, if the 2-way link latency is 1 ms (which is on the low side), the channel BW is 20 mbps and the packet size 600 bytes, then it takes only 0.24 ms for the SU to actually transmit the packet while it takes 1 ms to complete the channel grant/release signaling, which leads to a channel efficiency of only 0.24/1*100 = 24% ! Another way to look at this is in terms of available upstream bandwidth, which is only about 4.8 mbps (600 bytes every 1 ms), even though the channel has a raw capacity of 20 mbps.
    The TDMA system on the other hand has a channel efficiency that approaches the available link capacity. This is due to the fact that all transmissions are placed right next to each other (made possible by TDMA clock synchronization technology), and there are no large durations of "dead time" in between transmissions as in polling systems. There is some overhead in TDMA due to guardbands in-between bursts, but it is typically of a few microseconds in duration.
  • Upstream Latency: The upstream latency in polling systems is proportional to the number of SUs that are active. Polling systems also have problems distinguishing between active and in-active SUs which further aggravates this problem since the BSU may end up polling SUs that don't have any data to transmit. In a poorly designed polling system, this may lead to the situation that the upstream latency is proportional to the total number of SUs on the channel. In general even well designed polling systems are forced to expend link resources in polling idle SUs, since there is no way an idle SU can transmit upstream without being polled first. They try to reduce the overhead by polling idle SUs less often, but this leads to higher upstream latency for the first transmission.
    TDMA systems use a mixture of contention and piggybacking mechanisms for upstream transmissions. If a SU has data to transmit it sends a REQuest upstream in a contention slot (see Figure 2, typically multiple contention slots are provided per frame, and they are much smaller than data slots to conserve bandwidth). If there is a collision, then the SU backs off and re-transmits. Once the first transmission gets through, all subsequent transmissions are in the piggybacked mode. The resulting upstream latency is of the order of 2 to 3 link round trip times, and also does not blow up with increasing number of SUs. Also note that idle SUs do not consume any link resources in TDMA systems.
  • Bandwidth Management and QoS Support: Managing upstream bandwidth and assigning guaranteed bit rates to individual SUs is difficult to do in a polling system. This is due to the fact that the only control that the scheduler in the BSU has over bandwidth assignments is the frequency of grants that he gives to each SU. The scheduler can assign more bandwidth to an SU by giving it more grants, but this is a very crude control at best and ends up increasing the system latency and penalizing other SUs. Polling systems are incapable to supporting real time traffic such as voice and video, since the interval between polling instants is determined by the number of SUs, rather than by real time constraints of the traffic itself.
    TDMA systems on the other hand can use sophisticated scheduling with advanced bandwidth management. The Aperto® scheduler for example, used weighted fair queueing for scheduling upstream transmissions, which gives the service provider very fine control over the amount of bandwidth allocated to each SU.