Yuliang Li

Hi, These 2 parameters are independent. If simulation stop time is earlier than t, simulations will stop without running all flows (since some flows are indicated to start after simulation...

Yes, simulation is slow :) In my experience, running 320 hosts, at load 0.3 (or 0.5?), it takes 1 day to simulate 0.1s~0.2s. But your number seems too slow to...

I think you can capture the packet trace, and then calculate the throughput yourself. That's what I did :) To capture packet trace, you can check https://github.com/alibaba-edu/High-Precision-Congestion-Control/blob/master/simulation/run.py#L76 With this option...

Yes, you can check the paper Figure 11. On Sun, May 1, 2022 at 5:45 AM 0-polarstar ***@***.***> wrote: > 我试过（DCQCN + enable win），PFC没有触发。但只有*DCQCN*才会触发 PFC。合理吗？👀 > > 那么PFC触发时的测试用例是什么，测试的规模是多少？我已经尝试了 websearch 30%load...

Hi, if you have 9 ToR * 16 server/ToR = 144 servers, how do you get 1000-scale incast?

The incast load is 2% of the network capacity, which means incast continuously appear in the network, not just one time. What I did was calculating the average inter incast...

dcqcn and dcqcn_vwin were used in the HPCC paper. On Fri, May 6, 2022 at 10:20 PM Yuliang Li ***@***.***> wrote: > The incast load is 2% of the network...

How long did you simulate? To achieve 2% load, on average the number of incast per second is: 320 (hosts) * 100Gbps * 0.02 * / (500KB*60) = 2667.

Hi, dctcp needs has_win set to 1. You can check this line: https://github.com/alibaba-edu/High-Precision-Congestion-Control/blob/master/simulation/run.py#L131

Thank you. Could you please provide more details on this? The call chain you described seems legitimate to me, as after sending one packet, switch buffer occupancy decreases, so switch...