Measure the latency for a specific throughput/datarate method


Hi,

I would like to measure the latency for a specific
throughput/datarate. With the "ping" command it's possible to define
the interval between packets, and the size, but if i do the following
experiment:
datarate wished: 5 mbps
packet size: 1470 Bytes
interval between packets= i

i=1470*8/5000000=0.002352 sec

and then I run to my own network card:
sudo ping 10.1.160.46 -s 1462 -i 0.02352
i got:
>> --- 10.1.160.46 ping statistics ---
78 packets transmitted, 78 received, 0% packet loss, time 826ms
rtt min/avg/max/mdev = 0.020/0.027/0.049/0.009 ms

There were 78 packets transmitted in 826 ms, so
78*1470*8/0.826=1110508.4 bps =1.11 mbps
Is not what was expected, what is the parameter i'm forgetting to have
in mind to get the right result?

Because of this i decided to use this method:

   > iperf -c 192.168.101.11 -u -b 5m -t 30
and run simultaneously the ping command with a low datarate:
sudo ping 192.168.101.11 -s 1462 -i 1



so the packet traffic added by the ping command is insignificant when compared to the 5mbps added by iperf, and i'm sure that i've a flow with 5mbps.

What do you think about the method, is there any problem?

By the way, all the results i got were the same for all flow
datarates, when the values were lower than the channel bandwidth, but
when the value was a little bit higher, the results were:
tiago --at-- ksat67:~$ sudo ping 192.168.101.11 -s 1462 -i 1
PING 192.168.101.11 (192.168.101.11) 1462(1490) bytes of data.
1470 bytes from 192.168.101.11: icmp_seq=1 ttl=64 time=936 ms
1470 bytes from 192.168.101.11: icmp_seq=2 ttl=64 time=1253 ms
1470 bytes from 192.168.101.11: icmp_seq=3 ttl=64 time=1564 ms
1470 bytes from 192.168.101.11: icmp_seq=4 ttl=64 time=1840 ms
1470 bytes from 192.168.101.11: icmp_seq=5 ttl=64 time=2175 ms
1470 bytes from 192.168.101.11: icmp_seq=6 ttl=64 time=2481 ms
1470 bytes from 192.168.101.11: icmp_seq=7 ttl=64 time=2567 ms

The value of the round-trip-delay was always increasing, and the flow
was crossing 2 network nodes, so can i conclude that the delay
increases because the buffers are getting full?

thanks, Tiago Junqueira



Other Mailing lists | Author Index | Date Index | Subject Index | Thread Index