curtisb,
We got off on the wrong foot here. My initial responses to your contribution were in complete and enthusiastic agreement! Heck, I wholeheartedly endorse everything you've said about jumbo frames subsequently, it's just orthogonal to my very limited point.
I honestly never thought my parenthetical aside which was intended to obliquely throw some shade over someone else's claims would ever become such a source of contention. I didn't think it was even mildly controversial, and I said it the way I did to
avoid a direct confrontation. Basically, instead of shouting that 950 was BS, I was trying to say that you can't actually achieve that with normally wired internet, but since someone with jumbo frames theoretically could, I didn't want to dispute the claim with a false absolute. Hence I modified it to include that possibility and off to the races we went!
Again, totally not my intention and I did not and do not suggest that anyone actually use jumbo frames in their regular gigabit ethernet. That's a very bad idea for the numerous and entirely correct reasons that you have very appropriately elaborated upon!
---
Here's what I am actually saying, thoroughly written out:
1) Empirically, anyone with normal wired gigabit ethernet, inter-mediated by a switch or not, shouldn't expect to see bandwidth better than 940mbps +/- ~3mbps. This my recurrent finding with iperf in numerous scenarios as well as what is generally discussed on the internet whenever someone asks why they don't actually get 1000 mbps with their gigabit.
Anyone can verify this themselves.
Now, can you do (very, very slightly) better with jumbo frames? OK, I admit it: I don't actually know that empirically as I don't think I've ever done that. As you have said, there isn't any point! I mean, I agree totally, we are talking of a theoretical "improvement" by scant mbps an abstract, entirely artificial and very unrealistic benchmark at the cost of unrelentingly brutal practical problems starting with the extremely high probability that your equipment isn't even compatible (i.e.
you can't even do it in the first place!).
As I have said repeatedly, I 100% absolutely and totally agree completely!2) Theoretically, as I said before, we can math it out: add up packet overhead + frame overhead per mtu, and get a percentage, rigtht?
Out of a 1500 MTU size, you have the 20 bytes of the IP header, 20 bytes of the TCP header for the packet overhead.
https://en.wikipedia.org/wiki/IPv4#Headerhttps://en.wikipedia.org/wiki/Transmiss ... _structureThat's 40 bytes.
For the ethernet frame, you have 7 bytes of preamble, 1 byte of frame delimiter, 6 byes of MAC destination, 6 bytes of MAC source, 2 bytes of the type, then we have our 1500 packet payload, then another 4 bytes of CRC, and then 12 bytes of frame spacing.
https://en.wikipedia.org/wiki/Ethernet_frame#StructureThat's another 38 bytes.
So we have 78 bytes.
1500 - 78 = 1422
1422/1500 = .948 = 94.8%
Thus, theoretically speaking, the maximum you could -ever- reach over gigabit internet (with TCP and 1500 MTU) would be 948mbps.
Realistically, you're not going to quite hit that on the nose, hence the empirical observation of like ~940mbps.
And, with jumbo frames of 9000, it's 8922 / 9000 = .991 = 99.1%
The above is the only thing I was ever trying to say: that with jumbo frames, yes, theoretically you can beat 950 mbps.
Again, I just didn't want to say something that was a false absolute, so I qualified it.
----
Hence, my suspicion of that number. 950mbps is nice and round, but it isn't real. And if you are seeing it across a powerline network, how come I can't see it when I have host-to-host 1 meter CAT 6 connection?
It just doesn't make any sense, especially since you are using regular ethernet to get to that powerline networking link on both sides in the first place.