Dave Oxley <dave <at> daveoxley.co.uk> writes:
> Yeah, its a hub. I'll stop trying to set it to full duplex now. > It is 40 times quicker in one direction than the other. Can you > give me a hint where to go from here. Try connecting the 2 systems with only a 'cross-over cable' run the applications and make measurements. If this results in an increase in the bandwidth in either direction, you may want to put systems back on the hub, and have a third system run ethereal. Look at your data traffic and see if anythingelse is using the bandwidth from either of these 2 system or what else is plug into the hub/switch. Is the hub a 10Mbps only hub/switch, check that. On 10 Mbps ethernet hubs,you can never reach the full 10 Mbps, in fact with many systems chattering,the practical throughput is marginally around 33%. If when you are on the cross over cable and you get similar poor results,then the problem may be in the ethernet driver code, kernel, irq settings or some other low level part of the kernel/modules, especially if you get the same skewed results with several different applications moving data between the systems. But, if when you move data between these 2 isolated system, and get different bandwidth performance semantics, then the problem is most likely between the applications or a bottleneck in the application code (poor data structure for example). Make sure you computers are not resource limited, thus blocking the processthat you are running to move the data. Top and ntop are just a few toolsto help track down these sort of issues. Sadly, you may have a complex mix of part or all of these aforementioned issues... hth, James -- gentoo-user@gentoo.org mailing list
