UNIX for Dummies Questions & Answers

I am trying to determing how long it will take to transfer 384 Gb of data across a 100 Mb full ethernet. If I am correct, I come up with 36 Gb per hour. Surely that is not correct. I assumed 100 megabit per second is 10 megabyte per second, which is 600 megabytes per minute and 36 GB per hour. Where did I go wrong?
Thanks

You did not take into account the underlying protocol overhead, including TCP/IP and the application protocol. Also, on an ethernet segment, only one sender (source, destination) may talk at a time, which adds further overhead to the protocol overhead.

Plus, you did not take into account bus, OS and filesystem readwrite latency on both ends. The list goes on and on. It not possible to do a simple math calculation to determine the transfer rate of the actual data.

If you have a superfast system where the bus, IO, disk latency were insignificant, you would still have significant TCP/IP overhead.

I'm not sure you did go wrong. 100 megabits per second is actually 12.5 megabyte per second which is 750 megabytes per second and 45 gigabytes per hour. Your numbers have a better shot at being realistic due to all the stuff Neo mentioned.

The real way to do this is to send a GB across a few times and average the result to get an average time per GB.

Agreed. The realistic way to get a 'guesstimate' is to actually do a test and extrapolate upward. TCP segment size, MTU etc. all make a difference..... lots of variables.

When I used to do these tests for clients in NYC, I used a command-line program called: tcpspray.

Some useful links:

http://www.mnis.fr/home/linux/appli/.../tcpspray.html


http://www.slac.stanford.edu/xorg/nmtf/nmtf-tools.html