ndn-sci

Comments from the seminar:

1. Look at CPU utilization
2. Change ndncatchunks/putchunks timeout from 200 ms.
3. How much delay is due to queueing at NFD vs queueing at catchunks?
4. Make sure CS is at capacity - performance should be even lower
5. Multi-hop performance should be compared to multihop tcp.

susmit shannigrahi wrote:

Comments from the seminar:

1. Look at CPU utilization
2. Change ndncatchunks/putchunks timeout from 200 ms.
3. How much delay is due to queueing at NFD vs queueing at catchunks?

I think the queuing is rather at putchunks (where data packets are) than catchunks.

1. Make sure CS is at capacity - performance should be even lower
2. Multi-hop performance should be compared to multihop tcp.

Also look at Multihop UDP. + Figure out how to use larger packets over UDP (using NDNLP fragmentation).

For slide 10 "Performance over TCP tunnels":

• Can you retry with congestion reaction disabled?

Specifically for slide 11 "Performance over UDP tunnel":

• What is the packet loss?
• What are the measured RTTs? (min/avg/max)
• What is the performance of 1 UDP tunnel vs. 2 UDP tunnels?
• How many congestion marks do you get?
• Compare results with and without congestion marking.