Hi everybody !

I had tested his morning "srr-queue bandwith limit" command between 2 Windows XP workstation, thoose workstation were directly connected to the 2960G catalyst switch, I had use speed 100 command on both access port connected to my each workstation.

I had transfers from host 1 to host 2 one ISO file 629Mo. without any bandwith command specified the transfers was take 1 min. and 5 seconds, with bandwith command speficied on destination host the same transfer has took 5 minutes and 45 seconds.... which reprensent a connection near 14,59Mbps I had hope a connection around 50Mbps which would take around 2 minutes and 10 seconds (inluding IP overhead) if I'm base on 100Mbps transfers time

Someone can provide me an answer between observe results and which one I hope ?

Just to confirm my understanding, you connected two Windows XP hosts to copper gig 2960G ports configured for 100 Mbps (speed 100) and saw a transfer time of 1' 5", but when you configured both of these ports with "srr-queue bandwidth limit 50", instead of the transfer time doubling, it took about 5x as long? No other config changes were made and there were no other active ports?

If that's what happened, don't have a good answer for the difference in performance with just the information provided. The "srr-queue bandwidth limit" isn't precise, but wouldn't expect that much of a difference. Of course, there are many other possible variables especially depending on the nature of the data transfer.

Hi !

You had correctly understood, nothing was connected to this switch, and the same file was transfers between thoses 2 stations, the transmitting and receiving station was'nt changed between thoses 2 transfers tests.

The data I had transfers is an ISO image CD files.... I'like to know what this can make differences the type of data was transfers if it is always the same data was transfers in same direction each time !

Thanks a lot !

The data content shouldn't make a difference, file I/O on either system can make a difference. Which is why to obtain more accurate network performance some kind of traffic generator is used (to elimimate file I/O as a performance variable).

Ok,

but what about of the difference between expected result and the result I had with trafic shapping on this switch, is exist other solution more precise than bandwith limit command I can use on this plate-form ?

As I already wrote, ". . . wouldn't expect that much of a difference.", however, again, other variables could be involved. I've touched upon host file I/O, but there are others. (For example, two more that come to mind, actual IOS version being used and actual device configuration.)

Besides the port "shaping", you might also try SRR queue shaping.