search

gkvraman / Congestion-Control-Algorithms-Using-DQN

4 stars 0 forks source link

A problem about new_ssThresh #2

Open Cheryl0605 opened 4 years ago

Cheryl0605 commented 4 years ago

I found that when training rl-tcp in ns3gym, according to the definition, action = [new_ssThresh, new_cWnd], but there was no updated threshold in the training log for each step. I haven't found the problem. Looking forward to your help. Thanks.

gkvraman commented 4 years ago

Hello, I have used the template provided by rl-tcp. The algorithm which I used does not need new_ssThresh. It only focusses on present state action which is adjusting cwnd. Hence I did not use new_ssThresh.

On Thu, 9 Apr, 2020, 7:40 pm Cheryl0605, notifications@github.com wrote:

I found that when training rl-tcp in ns3gym, according to the definition, action = [new_ssThresh, new_cWnd], but there was no updated threshold in the training log for each step. I haven't found the problem. Looking forward to your help. Thanks.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/gkvraman/Congestion-Control-Algorithms-Using-DQN/issues/2, or unsubscribe https://github.com/notifications/unsubscribe-auth/AK6426TWJLN4O4IIFU7KQPTRLXJNXANCNFSM4MEYJ3YQ .

Cheryl0605 commented 4 years ago

Thank you very much for your reply, and I have another problem. I have run your code on ns3gym, and the maximum throughput is about 210,000 bytes/s. However, in the same network topology and bandwidth and latency configuration, NewReno can reach 250,000bytes/s. In this method, is 210,000 bytes/s the maximum limit of throughput? Or is there any way to continue to improve this throughput? Looking forward to your help. Thanks.