1. Native Simulation of Round-Robin Queuing
Lachlan Richardson
Glen Summers

2. What is Round-Robin Queuing?
Each queue takes turns being served
Remember previous Queue served
Circular servicing
Max-min fairness -> small or large packets, doesn’t matter. Each get their own service time?!?!?!

3. Context Used in Data Packet Scheduling Multiple incoming paths
Effective utilization of network capacity
‘Share the load’
Equal priority for packets
Possible scenario:
New data channel built
Requires another server, OR Round-Robin-Queuing
Use existing server to process new data channel
Split server processing equally

4. Simulation Setup Native discrete time simulation Fixed packet size
Fixed server service rate
1 million packets per simulation
Vary:
Max packets in queue
0 <= Pr(pkts arrival) <= 1
Correlative probabilities between queues
Measure:
Dropped packets
Draw discrete time diagram on the board
OBJECTIVES:
Simulate round-robin queuing
Extrapolate results
Find relationships

5. Intra-Queue probabilities equal, .3<P<.5
Same % arrival to both queues
Vary Probability
3 approach 7 queue limit, these are before 50/50 limit
Converge: towards 0 as queue limit increases

6. Intra-Queue probabilities equal, .45<P<.55
Around 50/50 line
Stable/unstable

7. Intra-Queue probabilities equal, .6<P<.9
% above sum 50/50 (=1)
Remainder = constant value. EG
60 – 50 = 10 (GREEN LINE)
Converges to const. value

8. Varying queue probabilities, Sum(P) = 1
Interesting -> closely related
Approach 1% pkt 18 queue limit
CONVERGES

9. Varying queue probabilities, Sum(P) = .9
Biggest difference = least loss -> BLUE!!! 85/05
GREEN = WORST,

10. Conclusion
Sum of probabilities should be kept less than 1 for a stable system
Having a large difference in probabilities between 2 queues is not a concern
Increasing queue limit decreases packet loss exponentially, however would increase the time packets spend in the system
Adding a second queue to a server with round robin scheduling works well as long as the P of arrival to each queue is kept less than .5
As the P of arrivals approach .5, ensure there is adequate queue lengths
Largely different probabilities between 2 queues:
Stability increases!!!!