chuanyuxue / udp-latency Goto Github PK

Hi @ChuanyuXue,

I got similar results when testing udp_rrt and udp_latency. It seems like division 2 of udp_rrt.py needs to be removed in the following function.

def listen(self, buffer_size: int, verbose: bool, save: Optional[str], q: Queue)-> None:
        latency = 0.0
        while True:
            msg, _ = self._udp_socket.recvfrom(buffer_size)
            recv_time = time.time_ns()
            packet_index = int.from_bytes(msg[:4], 'big')
            send_time = int.from_bytes(msg[4:12], 'big')
            old_latency = latency
            latency = round(((recv_time - send_time) * 1e-9) / 2, 6)
            jitter = abs(latency - old_latency)
            recv_size = len(msg)

Hi @ChuanyuXue, a quick question is what is the reason we choose socket instead of scapy to implement the udp latency? Thank you very much for your answering.

When I run it on a single Windows computer, I get a divide by zero with the default options:

C:\Users\Devin\Downloads>C:\Python310\python.exe udp_latency.py -c
| ---------- Sychonizing Server & Client by PTP ------------ |
| Client: 20002 | Packet: 1 | Time: 1716832507661762600 | Data size: 1468 |
Traceback (most recent call last):
File "C:\Users\Devin\Downloads\udp_latency.py", line 269, in
client.send(
File "C:\Users\Devin\Downloads\udp_latency.py", line 102, in send
* (len(self.log) / (frequency * (current_time - start_time) * 1e-9))
ZeroDivisionError: float division by zero

I can run it fine by using --dyna False on the client side.

When I run it on my computer, the current_time and start_time are the same, which results in a divide by zero:
frequency: 1.0
current_time: 1716832724571958100
start_time: 1716832724571958100
current_time - start_time: 0

I suspect that this is only an issue on Windows and is because of the following:
https://stackoverflow.com/a/1938096

I wonder what the best approach is to prevent the divide by zero, would something like the following work or would this skew the results?
if (current_time == start_time):
current_time = start_time + 1 # prevent divide by zero if using dyna on Windows due to +/- 16 ms accuracy (https://stackoverflow.com/a/1938096)

Would it be better to run without dyna?