I did a stress test, when trying 4096 connections, the lost is very high, above 50%

[archer]

Hi,

Is there anything I have missed caused such high lost, or is it too much loaded for 4096 connections?
Can anyone kindly help? Thanks.

test environment

- two machines A and B, both have 8 cpus and 8GB mem

- A,B connected in LAN max bw 500Mbps (tested with iperf)

turnserver runs in A, syslog disabled
turnserver -z 

turnutil_uclient runs in B

turnutils_uclient -y -l 75 -m 4096 -n 5000 <address-of-A>

result

output of turnuils_uclient shows there are lots of lost:

235: start_mclient: msz=395, tot_send_msgs=5120000, tot_recv_msgs=2229318, tot_send_bytes ~ 384000000, tot_recv_bytes ~ 167198850

236: start_mclient: msz=203, tot_send_msgs=5120000, tot_recv_msgs=2229318, tot_send_bytes ~ 384000000, tot_recv_bytes ~ 167198850

237: start_mclient: tot_send_msgs=5120000, tot_recv_msgs=2229318

237: start_mclient: tot_send_bytes ~ 384000000, tot_recv_bytes ~ 167198850

237: Total transmit time is 116

237: Total lost packets 2890682 (56.458633%), total send dropped 0 (0.000000%)

237: Average round trip delay 10.271851 ms; min = 0 ms, max = 158 ms

237: Average jitter 1.596343 ms; min = 0 ms, max = 121 ms

iperf result:

I tested the network with:

iperf -u -c <address-of-A> -p 5001 -b 500M -t 10 -i 3 -d

it looks okay:

------------------------------------------------------------

Server listening on UDP port 5001

Receiving 1470 byte datagrams

UDP buffer size:  208 KByte (default)

------------------------------------------------------------

------------------------------------------------------------

Client connecting to <address-of-A>, UDP port 5001

Sending 1470 byte datagrams, IPG target: 22.43 us (kalman adjust)

UDP buffer size:  208 KByte (default)

------------------------------------------------------------

[  4] local <address-of-B> port 26102 connected with <address-of-A> port 5001 (peer 2.0.10-alpha)

[  3] local <address-of-B> port 5001 connected with <address-of-A> port 6172

[ ID] Interval       Transfer     Bandwidth

[  4]  0.0- 3.0 sec   188 MBytes   524 Mbits/sec

[  3]  0.0- 3.0 sec   188 MBytes   524 Mbits/sec   0.023 ms    2/133761 (0.0015%)

[  3] 0.00-3.00 sec  19080 datagrams received out-of-order

[  4]  3.0- 6.0 sec   187 MBytes   524 Mbits/sec

[  3]  3.0- 6.0 sec   187 MBytes   524 Mbits/sec   0.028 ms    2/133744 (0.0015%)

[  3] 3.00-6.00 sec  383 datagrams received out-of-order

[  4]  6.0- 9.0 sec   188 MBytes   524 Mbits/sec

[  3]  6.0- 9.0 sec   188 MBytes   524 Mbits/sec   0.017 ms    0/133751 (0%)

[  3] 6.00-9.00 sec  2378 datagrams received out-of-order

[  4]  0.0-10.0 sec   625 MBytes   524 Mbits/sec

[  4] Sent 445821 datagrams

[  3]  0.0-10.0 sec   625 MBytes   524 Mbits/sec   0.018 ms   23/445821 (0.0052%)

[  3] 0.00-10.00 sec  31091 datagrams received out-of-order

[SUM]  0.0-10.0 sec   812 MBytes   682 Mbits/sec   0.023 ms   25/579582 (0.0043%)

[SUM] 0.00-10.00 sec  50171 datagrams received out-of-order

[  4] Server Report:

[  4]  0.0-10.0 sec   625 MBytes   524 Mbits/sec   0.000 ms   21/445821 (0%)

[  4] 0.00-10.00 sec  1960 datagrams received out-of-order 

[Oleg Sevryukov]

Check packages TTL size, it could be the reason of packages lost

вторник, 31 октября 2017 г., 6:57:40 UTC+2 пользователь archer написал: