Virtual output queueing

Virtual Output Queues (VOQ) are an input queuing strategy for use in telecommunications and computer network switches. It addresses a common problem known as Head-of-line blocking.^[1]

Description

In VOQ each input port maintains a separate queue for each output port. It has been shown that VOQ can achieve 100% throughput performance with an effective scheduling algorithm. This scheduling algorithm should be able to provide a high speed mapping of packets from inputs to outputs on a cycle-to-cycle basis. VOQ mechanism provides throughput at a much higher rate than the crossbar switches without it.

For example, consider a 2x2 crossbar switch.

     -------- 
a--->|-\--/-|--->--0
     |  \/  |
     |  /\  |
b--->| /--\ |---->-1
     --------

Lets say that data "0" arriving at port a or b will go to output port 0 Similarly data "1" arriving at port a or b will go to output port 1

So the number of combinations that can happen at the input port a, b are

If data at the input is "00", this means both the input data at time t are contending for the same output port 0 and the output port 0 can't route both the data at the same time as it can handle unit data per time slot. So in this case the efficiency of the 2x2 switch (without VOQ) is 0.5

Same is the case for data "11" in which the efficiency is 0.5

Similarly for data "01" and "10" the efficiency is 1 as there is no contention as both the data go to both output ports 0 and 1.

Since it's a 2x2 switch, the probability that anyone of out of these 4 combinations of data will occur = 0.25

So the efficiency of this 2x2 switch is

  (0.25 * 0.5) --> for data 00 
+ (0.25 * 0.5) --> for data 11
+ (0.25 * 1.0) --> for data 01
+ (0.25 * 1.0) --> for data 10
---------------
 = 0.75 (75%)

So we see that the 2x2 crossbar switch is working at 75% efficiency to give out data from input to output.

As n increases, for nxn switches this causes further degradation in efficiency.

VOQ (Virtual output queing) overcomes this problem by introducing extra buffers (queues) per port.

Lets revisit the same scenario with 2x2 switch with VOQ support.

     -------- 
a--->|-\--/-|-OO-->--0
     |  \/  |
     |  /\  |
b--->| /--\ |-OO--->-1
     --------

Here each output port has n buffers per port to accommodate maximum no of simultaneous data which each port might receive at a time.

So this buffering mechanism removes the bottleneck per port on peak time and distributes it over a period of time increasing the switch performance.

There are many algorithms for design and implementation of fast VOQ. Nick McKeown and a group at Stanford University for example published a design in 1997.^[2]

References

^ Mark W. Goudreau; Stavros G. Kolliopoulos; Satish B. Rao (2000). "Scheduling Algorithms for Input-Queued Switches: Randomized Techniques and Experimental Evaluation". Proceedings of IEEE INFOCOM: 1634–1643. doi:10.1109/INFCOM.2000.832562.
^ Nick McKeown; Martin Izzard; Adisak Mekkittikul; Bill Ellersick; Mark Horowitz (1997). "Tiny Tera: a packet switch core" (PDF). IEEE Micro: 26–33. doi:10.1109/40.566194.

This computer science article is a stub. You can help Wikipedia by expanding it.

[1] Mark W. Goudreau; Stavros G. Kolliopoulos; Satish B. Rao (2000). "Scheduling Algorithms for Input-Queued Switches: Randomized Techniques and Experimental Evaluation". Proceedings of IEEE INFOCOM: 1634–1643. doi:10.1109/INFCOM.2000.832562.

[2] Nick McKeown; Martin Izzard; Adisak Mekkittikul; Bill Ellersick; Mark Horowitz (1997). "Tiny Tera: a packet switch core" (PDF). IEEE Micro: 26–33. doi:10.1109/40.566194.

[1]

[2]