what makes gigabit ethernet so fast?
 

Can some one explain the technical difference between gigabit ethernet (over copper wiring) which allows it to be so much faster than, say, 100 Mb ethernet? Does it just transmit signals faster? Or does it transmit multiple signals at once? Or...? I read that the gigabit standard requires use of cat 6 cable, which is rated for 250 Mhz bandwidth, but cat 5 cable is rated for 100 Mhz bandwith -- obviously not a 10 fold increase there.

Here's a (very) detailed article on the topic:
http://www.siemon.com/uk/white_papers/04-01-15_cat6.asp
In summary:
Higher quality cable, higher signal to noise ratio, and higher bandwidth.

In addition, 1000BaseT uses all 4 pairs vs 10/100 which only uses 2 pairs. The adapter hardware is different, 100baseT uses 3 differential voltages, 1000BaseT uses 5. More twists per inch also reduces crosstalk and improves throughput.

A car going 100 miles an hour versus a jet going 1000 mph.

The occupants are 1's and 0's in the ethernet case.

Please bare in mind that a 1 Gbps will not increase real throughput by 10 over 100 Mbps, max throughput would be only 4 times faster !

Why?

Unless you're saying that 10/100 is faster than 10/100, then this is incorrect. I personally get well over 4x 10/100 speeds with cheapo consumer-grade gigabit hardware.

10/100 is not faster than 10/100 , it is equal !
I was saying that throughput is 40-50 MB/s on 1000 base versus 12 MB/s 100 base. Many people tend to expect exact ratio as speed increase !

My 2 Dell's PE1950 connected via CAT 6 1 meter cable achieve somewhere in 43-52 MB/s

Just because you only get 43-52 does not mean that's what everybody gets...I typically get about 60 MB/s encrypted transfer speeds on my network, and 90-105 MB/s unencrypted, and that's using cheap consumer gigabit hardware (dlink, linksys, etc).

Here's a 10GB dd dump over an NFS mount:
And that's on a pretty active network, with two pretty active machines (load average of 5 on one (8 core), 3 on the other (12 core))

Theory and real world are two different sales pitches.

My car could do 186. In real world, I can barely get it to 70 before I get a ticket.

Ever driven a Suzuki (car not motorcycle) ... you'll have a hard time getting a speeding ticket. I drove one for my driving exam, and maybe that's why they bought these for this purpose. Certainly I couldn't get it past 80-90 mph depending on wind.

Unlike a road, there are no speed limits inside a twisted-pair cable, and unlike a car, an Ethernet tranceiver is running full throttle all the time. A car analogy is perhaps not very appropriate in this case.

Unlike with wireless technologies, a working 1000Base-T connection is capable of a throughput of 1 gigabit/second in both directions, no ifs or buts.

@yo8rxp: The poor throughput you're seeing is probably due to the equipment not being able to fill the pipe to capacity. You should be seeing an effective speed of 1 gigabit/second, unless the tranceivers are broken or the NICs buggy (certain RTL chipsets comes to mind). Another possibility is that the connection could be running in half-duplex mode, but that's really quite unlikely.

well , my bad !yes indeed , using ssd drives is one , old drives is another issue, my mistake was to face here real world using old SAS (94 MB/s read /50 write on my drives )
I will have to reconsider changing drives in my main server / back up server , but unfortunatley , they are expensive ones <!-- price per /GB -->

True ! i did not ever run a test against ram drive /dev/shm or /dev/zero towards server, there should be a relevant test limited only br RAM speed

And nobody will issue speed tickets here for high speeds
Cheers !