When working with the open source D-ITG traffic simulation software, we are running into traffic generation limitations. While the ultimate goal is to maximize the traffic generation potential, we must at least be able to determine what the maximum is for a given platform. So these are the two goals of this project: determine the potential of a given platform, optimize the traffic generation process to raise the existing ceiling.
The limitations we see appear to be based on packet rate. When we try to increase the total bandwidth sent by increasing the packet rate, ITGSend does not send the requested number of packets. When we write a local log file, we see this there. The maximum that it will send does vary with different hardware. It seems that the packet generation itself is causing the limitation. We would like to be able to run a series of test simulations to a loopback interface in order to determine the capabilities of the hardware we are running on. Currently, running ITGRecv on a system, then running ITGSend to 127.0.0.1 on the same system causes ITGRecv to crash.
The second part is increasing this limit. Again, since ITGSend is reporting that it sent fewer packets than requested, it seems that the problem is in packet creation. The simulations don't contain real data, of course, so if it is packet generation that is imposing the limit, and playing of a PCAP would be more efficient, then we would like to do that. For instance, if all of the voice calls use a 50B packet, we can have a PCAP of a 50kb packet and just replay it as needed for the stream. Of course, if there is a better (read more efficient) way to generate the traffic, we are happy to go that route.
This work is important to move forward with other projects and we would like to complete as quickly as possible.