I am presently pursuing my PhD in computer science. I have chosen the topic of my work to be an SDN based approach for data center traffic engineering. I have thought of a new topology for DCs and am thinking of comparing the performances with the existing fat tree topology, as I think other topologies like hyperX, hyperCube etc. won't work well in all the cases. I took the SDN course by Dr. Feamster. I need to know the parameters of a DC topology that really matters for comparison. Can I get the actual map of the scalability by creating topologies in mininet. How can I proceed in this?
Hello Nabajyoti,
You can define basic topology structure (3 layer clos, etc). in mininet by using aggregate node functionality to combine multiple links for more accurate modelling. Also instead of creating the whole DC topology, start of by working with fewer nodes with high inter-link bandwidth and generate traffic to get a feel of how it would work in a larger network. Once you identify the bottlenecks in your proposed topology vs. fat tree (for example), you can iron out algorithms for load balancing i.e. optimal path selection etc.
On a side note, instead of creating a whole new topology from scratch, it would be better to optimize existing fat-tree through novel traffic engineering solutions which are of high relevance and readily applicable to present DC needs.
Regards,
Taimur
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India