| 1 |
Harry Putnam <reader <at> newsguy.com> writes: |
| 2 |
|
| 3 |
|
| 4 |
|
| 5 |
> But still, when I'm trying to measure how much data is moving |
| 6 |
|
| 7 |
emerge bwmon, |
| 8 |
|
| 9 |
It measures across the ethernet ports, so adjust your test, |
| 10 |
according to what you want to measure, crossing the ethernet |
| 11 |
port on the target system. |
| 12 |
|
| 13 |
|
| 14 |
> and it seems quite slow for what is supposed to be a gigabyte network. |
| 15 |
|
| 16 |
Gigabit ethernet rarely runs full out constantly, something, (ram, cpu, |
| 17 |
interface, swith-latency....) mucks things up. Do not let your "copper" |
| 18 |
get to long either! |
| 19 |
|
| 20 |
|
| 21 |
> gigabyte switch |
| 22 |
> | | |
| 23 |
> | | |
| 24 |
> (192.168.0.9) h4 h5 (192.168.0.17) |
| 25 |
|
| 26 |
> But also if I should be expecting h4 h5 to be able to use GigaByte |
| 27 |
> transfer speeds. |
| 28 |
|
| 29 |
Some fraction say 50% is good, if it is copper, unless the systems |
| 30 |
are smoking "gaming" systems or of very high quality resources. |
| 31 |
|
| 32 |
emerge iotop (then rebuild kernel) |
| 33 |
|
| 34 |
Rebuild your kernel to use "iotop" and see what happens. You may need |
| 35 |
to tweak your kernel parameters to optimize IO operations..... |
| 36 |
(just gessing here).... |
| 37 |
|
| 38 |
|
| 39 |
hth, |
| 40 |
James |
Archives