On 08.01.2015 18:37, Linda Walsh wrote: > Phillip Susi wrote: > >>speed in Bytes varies by protocol. 1Gb-Base-T ethernet maxes out > >>at a theoretical 125MB/s - divisible by 8. But 10Gb ethernet maxes > >>out at 1000MB/s -- with 20% of its bandwidth going to protocol > >>overhead. > > > >I'm not aware of any additional overhead that 10Gb ethernet has over > >1Gb ethernet, > ---- > See kernel messages for a 10b-T ethernet. > > [ 21.224641] ixgbe 0000:05:00.0: PCI Express bandwidth of 32GT/s available > [ 21.224644] ixgbe 0000:05:00.0: (Speed:5.0GT/s, Width: x8, Encoding > Loss:20%) > > I don't recall a 20% encoding loss in 1Gb or 100Mb ethernet and the kernel > displays > no such messages for the slower speed cards. The message speaks about PCIe. So the 40GBit/s (a.k.a. 40GT/s) is in effect 32GBit/s on the PCIe side. 5.0 GT/s = PCIe Gen. 2. PCIe Gen. 1 & 2 use 8b/10b encoding. IOW for every 8 bits of payload 10 bits go ever the wire. This is 20% the enconding loss the message speaks about. PCIe Gen 3 (and 4) use an enhanced encoding called 128b/130b. IOW for every 128 Bit of data 130Bits are send over the wire, so only 8GT/s (instead of 10GT/s) were needed to (nearly) double the effective datarate in Gen 3. See: http://en.wikipedia.org/wiki/PCI_Express Which still leaves well enough headroom to get the 10Git/s for the ethernet-connection across. -- Matthias -- To unsubscribe from this list: send the line "unsubscribe util-linux" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html