Known iperf problems
This page concerns problems discovered with iperf. These tests were conducted with both version 1.2 and 1.6.1.
| Problem | Description | Version |
| Memory Limit | There is a physical limitation in the amount of memory that iperf can hold to calculate the throughput. It was found that this value is 4gbytes. This is equivalent to a duration of 32 seconds if we can transfer at GigE rates (1000mbit/sec). | 1.6.1 and below |
| Server window limit | There is a physical memory limit of 2mbytes settable for any value of the server window size. This value is shown in the reciever window size. This may impete the throughput along high latency paths. | 1.6.1 and below |
| Sender socket buffer size incorrectly set | The sender socket buffer is always (under linux 2.4.x) to a value twice that requested. | 1.6.1 and below |
| Server window incorrectly set | The server window is set to approximately 1.5 times the value requested. This is upto a maximum of that descibed in 'server window limit'. | 1.6.1 and below |
| Defining a number of bytes to send has an upper bound | If we use the '-n' option, the amount of data to be transferred is limited to 4gigabytes. if we define a value greater than 4gb, it wraps around. ie if we defin 4gb and 1 mb, it will only send 1mb. | |
| Defining a time to send reports an incorrect throughput value. | If we send for a time greater that results in sending more than 4gbytes, the throughput reported by iperf is incorrect. |
| Wed, 23 July, 2003 13:07 |   |
Room D14, High Energy Particle Physics, Dept. of Physics & Astronomy, UCL, Gower St, London, WC1E 6BT
