Iperf TCP throughput measurement result


Hi, All:
 
I am currently evaluating a high-performance radio (~300 Mbps) in the lab environment (short distance) using Iperf.  I am seeing different throughput results with 1 client and 5 clients.
 
Here is the one-way and two-way results shown below.

1.	One-way throughput

	*	Testing with one client

			*	------------------------------------------------------------
				Server listening on TCP port 5001
				TCP window size: 65.0 KByte
				------------------------------------------------------------
				------------------------------------------------------------
				Client connecting to 192.168.100.2, TCP port 5001
				TCP window size: 65.0 KByte (WARNING: requested 65.0 KByte)
				------------------------------------------------------------
				[  5] local 192.168.100.1 port 50475 connected with 192.168.100.2 port 5001
				[  5]  0.0-120.0 sec  1.99 GBytes    143 Mbits/sec
				[  4] local 192.168.100.1 port 5001 connected with 192.168.100.2 port 56734
				[  4]  0.0-120.0 sec  2.00 GBytes    143 Mbits/sec

	*	Testing with five cleints (-P5)

			*	 ------------------------------------------------------------
				Server listening on TCP port 5001
				TCP window size: 65.0 KByte
				------------------------------------------------------------
				------------------------------------------------------------
				Client connecting to 192.168.100.2, TCP port 5001
				TCP window size: 65.0 KByte (WARNING: requested 65.0 KByte)
				------------------------------------------------------------
				[  5] local 192.168.100.1 port 65130 connected with 192.168.100.2 port 5001
				[  3] local 192.168.100.1 port 50815 connected with 192.168.100.2 port 5001
				[  4] local 192.168.100.1 port 63251 connected with 192.168.100.2 port 5001
				[  9] local 192.168.100.1 port 64820 connected with 192.168.100.2 port 5001
				[  6] local 192.168.100.1 port 57045 connected with 192.168.100.2 port 5001
				[  3]  0.0-120.0 sec    631 MBytes  44.1 Mbits/sec
				[  4]  0.0-120.0 sec    652 MBytes  45.6 Mbits/sec
				[  6]  0.0-120.0 sec    649 MBytes  45.4 Mbits/sec
				[  9]  0.0-120.0 sec    649 MBytes  45.4 Mbits/sec
				[  5]  0.0-120.0 sec    639 MBytes  44.7 Mbits/sec
				[SUM]  0.0-120.0 sec  3.14 GBytes    225 Mbits/sec
				[  8] local 192.168.100.1 port 5001 connected with 192.168.100.2 port 64349
				[  3] local 192.168.100.1 port 5001 connected with 192.168.100.2 port 55116
				[  4] local 192.168.100.1 port 5001 connected with 192.168.100.2 port 51162
				[  5] local 192.168.100.1 port 5001 connected with 192.168.100.2 port 58124
				[  6] local 192.168.100.1 port 5001 connected with 192.168.100.2 port 50025
				[  3]  0.0-120.0 sec    647 MBytes  45.2 Mbits/sec
				[  8]  0.0-120.0 sec    655 MBytes  45.8 Mbits/sec
				[  5]  0.0-120.0 sec    640 MBytes  44.7 Mbits/sec
				[  4]  0.0-120.0 sec    656 MBytes  45.8 Mbits/sec
				[  6]  0.0-120.0 sec    616 MBytes  43.0 Mbits/sec
				[SUM]  0.0-120.0 sec  3.14 GBytes    225 Mbits/sec

My question is that what could cause the Iperf to have better throughput when 5 clients run at the same time.  One speculation is that the radio under evulation has dynamic TDD feature, which leads to a variable latency.  And the OS cannot cope well with a variable latency.  

Based on my knowledge,  even though it is fat pipe, I test the radios in very short distance.  I don't think it has anything to do with TCP tuning.

Any comments on this issue will be appreciated.

Jin Fan



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