Solved 125 MB/s on Gigabit Network

spanglefox

spanglefox

Just a quick question.

Recently I did a little optimising on a FreeBSD server and tested copying a 3GB file across the network, to a Windows machine. Windows' copy dialogue reported a burst speed of 122MB/s transfer speed before settling in a reported 100+MB/s area. Granted Windows' copy dialogue may not the most accurate of metrics.

This puzzled me, as I had understood that 'real life' transfer rates were deduced by dividing the connection speed by ten. (8 bits of data + 2 bits of start/stop signalling). Which would give a theoretical maximum speed of 100MB/s. So I Googled maximum transfer rates on Gigabit connections. Some reputable sites indicated that the maximum was indeed 125MB/s (1000 bits/8 bits).

Could anyone clarify if theoretical network speeds are divided by 8 or 10? It may be that I am still using outdated maths.
 
Your question is not really about theoretical network speed but it goes down to whether 1 byte is 8 or 10 bits. And answer for that question is 8.
 
There are indeed only 8 bits in a byte but due to protocol overhead estimations are closer when using 10. This has nothing to do with start or stop bits, they're only used on serial connections, not ethernet. But because packets needs to have MAC and IP addresses, port numbers and a few more bits for protocol features. So there's simply less room for the actual data that's being transferred.
 
Yes, that is where I realised that my calculations may have been from the "old school". I grew up using serial ports, dial-up modems and the like. The problem made me stop and check that I was not using outdated information.

I did wonder about overhead and I guess that is why "10" has worked for so long.

Thanks once again :)
 
I know this is already solved, but I wanted to add. I get about 116MB/s on Windows to Windows with SSDs. I think gigabit Ethernet is 8/10, but the raw signally rate I think is 1.25 Gb, giving a logical signal of 1 Gb. I know GPON has a 1.24 Gb/s up logical throughput, but uses frames based on time of 125 us, which gives a bit over 38K frames with much lower overhead. I think they use the same optics, just different raw signals.
 
You must log in or register to reply here.
Back
Top