| Number of double-values | Time (sec) |
| 1 | 0,000087 |
| 2 | 0,000091 |
| 4 | 0,000091 |
| 8 | 0,000096 |
| 16 | 0,000105 |
| 32 | 0,000123 |
| 64 | 0,000167 |
| 128 | 0,000251 |
| 256 | 0,000375 |
| 512 | 0,000551 |
| 1024 | 0,000908 |
| 2048 | 0,001605 |
| 4096 | 0,003016 |
| 8192 | 0,005899 |
| 16384 | 0,011814 |
| 32768 | 0,023258 |
| 65536 | 0,046097 |
|
||||||||||
|
||||||||||
As I'm not the only one using the machines I've tested my program on and I'm not aware of the
network parameters used, I can only guess the startup-time and bandwidth.
When looking at the results with a data-amount of 216 double values it can be derived, that the
machines the program was tested on, must be connected to a 100 Mbit network, as the results show
a maximum transfer rate of about 90 Mbit/s.
Looking at the results with a data-amount of one or two double-values it can be seen, that in this case the transfer-time does not really play a role. So startup-time must be in a range of 50µs and 100µs. Such low start-up times are necessary to make a large number of (short) consecutive transfers in a LAN possible.
The overall communication-time nearly doubles with a doubling of the amount of data transferred, what could in fact be expected.