Consider a software-defined network (SDN) with multiple physically interconnected SDN switches and an SDN controller. Attached is an illustration of the network. Multiple end hosts are connected to the switches. In the illustration, the blue drawings form the data-plane network, whereas the red drawings form the control-plane network. The switches may have different processing capacity. The end hosts communicate with each other through the network. The rates of the data traffic generated by the end hosts vary with time. Sometimes, the data rates may be high to potentially cause network congestion. Now, the question is, how the SDN controller controls the network such that the end-to-end packet delay from a host to another host is upper-bounded by a specified threshold?
Please propose your solution, as detailed as possible. The solution should use standard SDN protocols that commodity SDN controllers, switches support; the solution shouldn't add any software program to the end hosts.
Dear Sir,
First of all, to simplify consider end-to-end latency (T) is the sum of three components: T_prop (propagation), T_tx (transmission, processing...), T_buf (queuing).
On the one hand, T_prop can be easily minimized/guaranteed if you route your traffic over the shortest path (this is usually implemented by SDN controllers using a Dijkstra-like algorithm).
On the other hand, T_tx may be considered as a fixed value proportional to packet length and line rate.
Finally, T_buf depends on several factors, not only line rate or congestion, but also node architecture (output queuing, combined input-output queuin, and so on), and hardly can be ensured (not only in average but also in variation -jitter-).
As end-to-end latency is not easy to be theoretically modeled, there are some options you can choose to minimize the delay for specific flows:
1. Route traffic over shortest paths and pray for having limited congestion (operators usually dimension capacities to have a peak utilization of 60-70% of the capacity)
2. Apply QoS mechanisms like DiffServ to prioritize certain flows (i.e., high-priority flows are forwarded earlier than low-priority flows)
3. Apply active monitoring techniques (search for OWAMP/TWAMP techniques) and reroute traffic accordingly. These techniques are based on injection of background traffic to measure different metrics like RTT, throughput, and so on. Then, with this information the SDN controller can have a picture of the state of the network and then take decisions accordingly.
The funny thing here is how to implement options 2 and 3 inside the controllers and translate into the devices :)
Best regards,
José Luis
It's a nice question..how to set the threshold value, what is the criteria for setting this threshod value? With mininet design such topology, and set the link, capacity of the controller and switch. With iperf tool generate various flows. Use wireshark tool in background. Use any monitoring techniques to verify your traffic.
I believe a priority scheduling algorithm for multi level queues having different priorities would solve this problem.
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Waves