Emulation of Differentiated Services in VNUML CS 680

Emulation of Differentiated Services in VNUML CS 680 Class Project Course Instructor : Prof. Anirudha Sahoo Project Team members : Vijay Gabale (07305004) Sagar Bijwe (07305023) Manish Kumar (07405701)

Agenda Motivation Network set up in VNUML Architecture Implementation details Demo Experiments and Results Observations Future scope and Conclusion

Motivation Implementation of theoretical concepts into practice Get hands dirty with scheduling, token bucket, IP address prefix matching schemes Interesting questions What is order of improvement achieved by DS enabled service in terms of delay, jitter and throughput? How do real time flows behave in DS enabled network?

Architecture Classification Marking Drop Priority Client Network Router Core Network Boundary Router Client B Client A Core Network Core Router Client Network Router Scheduling Forwarding

Classification – Multi-Trie dst Approach 0 9 1 F 2 F 3 0 0 0 src 0 1 9 Destination trie 0 1 9 Source trie F 1 F 2 F 3

Drop Priority – Dual Token Bucket Two-Rate Three-Color Marker red yellow PIR/PBS Token bucket P CIR/CBS Token bucket C green

Experimental Set Up in VNUML client Boundary Router Customer Network Forwarding Router Core Network Server

What is delay improvement? Delay decreases considerably in DS enabled network for EF flow.

What is jitter varying? Jitter for DS enabled flow remains fairly constant.

How is throughput varying for two flows? Throughput for DS enabled flow was greater than disable flow.

What is the number of packets delivered for two flows? EF flow gets priority in terms of packet scheduling and delivery.

What is number of packets dropped for two flows? EF flow gets priority in terms of packet scheduling and delivery.

What is the effect of weight given to classes? (delay) Flow with more weight has lesser delay in DS enabled network.

What is the effect of weight given to classes? (throughput) Flow with more weight has more throughput in DS enabled network.

Observation The service given to one of two real time flows (EF) in the experiment was quite predictable. This is also in accordance with that Diff. Serv does not guarantee any end to end delay but provides sufficient resources to have improvement as compared to other flows. As we increased the token generation rate of one of the flows, more number of packets of the flow got admitted which in turn resulted lower number of packets getting dropped, lesser delay and higher throughput.

Conclusion Diff. Serv improves the performance of real time flows. (about 25% increase in terms of delay and throughput) The token bucket parameters, if set appropriately, are helpful in admission control.

References “An Architecture for Differentiated Services” RFC 2475 “Virtual Network User Mode Linux” “http: //www. dit. upm. es/vnumlwiki/”