Scenario: COP as server, SOLDIER as client, Ethernet 10Mbps with varied latencies. Messages of 500 bytes.
Protocol | 5ms | 30ms | 100ms | 300ms |
---|---|---|---|---|
TCPM | 3966 | 2900 | 2326 | 782 |
SCTP | 3366 | 2074 | 1570 | 882 |
Network latencies in ms, throughput in kbps (1kbps = 1000 bits/s).
Throughput falls as latency increases. It seems abnormal but it is caused by reception buffers smaller than "network latency X bandwith product". The buffers should have more room as latency increased. (Since most applications do not adapt their reception buffers, we preferred to use standard buffer sizes and see what happens).
As usual, SCTP was slower, but difference to TCPM fell as network latency was increased. Near to 300ms, SCTP surpassed TCPM.
Protocol | 5ms | 30ms | 100ms | 300ms |
---|---|---|---|---|
TCPM | 10303,5 | 60286,0 | 200266,0 | 600183,5 |
SCTP | 10355,0 | 60337,0 | 200298,5 | 600189,5 |
Network latencies in ms, transaction latencies in µs
SCTP latency is always bigger than TCPM's, but difference is irrelevant compared to the network latencies. Because of that, one of graphic lines is completely hidden, and the transaction latency is essentialy equal to network RTT (RTT = sum of network latencies at transmission and reception).