Gigabit LAN anyone??

[chez]

I have heard a lot of reports recently about Gigabit LAN only being slightly faster than 100Mbps. I read an article which was linked from a post on some forum or other which stated that, at best, you would get a 20-30% increase in speed and that consumer gigabit hardware was not very good.

Tonight I have had to copy 90Gb of data from one PC to another. I started it copying through my Netgear router and windows estimated it was going to take 178 minutes!

I thought I’d try gigabit out (as both PCs have onboard gigabit LAN) and whipped out my Cat5e crossover cable.

Suffice to say I saw one hell of an improvement, with windows estimating 50 minutes total (which turned out to be fairly accurate!).

My question is this – will the performance increase that I saw be maintained if I was to buy a Netgear gigabit switch or was such a boost only seen because I used a direct connection and was not using any of the ‘not very good’ hardware which I have seen mentioned…???

I hope this makes sense!

Thanks
Chez

[Rave]

You don't need a crossover cable for Gigabit m8. As for your question though, no idea, sorry.

Rich :¬)

Edit: The limitation in that 90GB file transfer was almost cerrtainly your hard disks sustained read/write performance.

[chez]

You don't need a crossover cable for Gigabit m8. As for your question though, no idea, sorry.

Surely you do if your router is only 10/100?

[javalord]

Surely you do if your router is only 10/100?

No, the mdi-X port can detect the signaling speeds and will adjust accordingly.

[nichomach]

chez (I assume) was using his two NICs directly wired to each other via a single cable. For that, crossover would be necessary as with any direct NIC-to-NIC connection. As far as signalling speeds go, a 10/100 switch (or one built into a router) will top out at 100Mbps. So if his router has a 10/100 switch built in, the only way to get gigabit speeds would be to unplug both machines from the router and connect them directly using a CAT5e cable (or a good quality CAT5). Rave's probably right as regards the limiting factor in the file transfer when that direct connection was in place.

As regards whether that performance would be maintained if a gigabit switch were used, there's no reason why not; a switch provides dedicated rather than shared bandwidth, so assuming the switch is something like a 4x10/100/1000, it should be able to cope with providing gigabit speed connections to all ports.

[rajagra]

90 GigaBYTES in 178 mins=8.4MBytes/s, near the upper limit for a 100MegaBIT/s connection, but slower than modern hard drives.
90GB in 50 min=30MB/s, typical limit for modern hard drives taking account inefficiencies of Windows file copying, but WELL within the limits for gigabit ethernet.
A gigabit switch introduces some latency over a simple cable connection, but won't slow things down noticably. Go for it!
(I tend to use B=bytes and b=bits but sadly this isn't standard, lol.)

[Rabs]

I'll do some tests using my gigabit hub and network cards asap. I saw an article on Toms where on average you get 30-40MB/s (yep Megabyte per second) over gigabit. As suggested your HD will probably be the bottle neck as its very unlikely it will reach those speeds sustained over small files although it may over a very large file.

My previous 10/100 network did 9.8MB-10MB second constant over a 25M cable, no reason for gigabit not to show a significant speed increase.

[Paul Adams]

I'll do some tests using my gigabit hub and network cards asap.

A very cool command-line tool for testing network speed and playing with MTU & RWIN sizes is Iperf:
http://dast.nlanr.net/Projects/Iperf/

e.g.
Say you have 2 PCs with IPs 10.0.0.1 and 10.0.0.2:
Test the connection speed one way...
On the 10.0.0.1 PC enter the command:
iperf -s -m
On the 10.0.0.2 PC enter the command:
iperf -c 10.0.0.1

Then test it the other way...
On the 10.0.0.2 PC enter the command:
iperf -s -m
On the 10.0.0.1 PC enter the command:
iperf -c 10.0.0.2


It's very handy as it's just emulating raw network connections, not file read/write operations, and ignores any overhead from the application layer.

[javalord]

If you can use a seperate interface for internet gateway and internal home network, you might want to use Jumbo frames on the Gigabit ethernet controllers coupled with a large recieve window for them. It's been shown in testing that those 2 combinations lower CPU util and improve throughoutput. But it cannot be done if your using the gigabit controller for internet too, as your net router may not like jumbo frames.

[Rabs]

UPDATE:

Tested the gigabit network with my rig (in sig) and my pals Gigabyte K8VNXP onboard gigabit. It averages about 30-33MB/s.

Only had chance to run a quick test 10GB or so but I'm pleased with the result. Gigabit network card for my server now

[Paul Adams]

Rabs, was that with a crossover cable or a gigabit switch (or was there no difference)?

Did you enable jumbo frames or play with the RWIN value at all?

Through my gigabit switch which doesn't support jumbo frames, Iperf reports my speed as around 300Mbps (37.5MiB/s), I seem to have misplaced all my crossover cables so I can't test directly between 2 of my machines right now (and see what a difference jumbo frames makes).

[Rabs]

Mine was through a Netgear gigabit switch, didn't have time to test different frame sizes etc. Just tried it on my pals too, that got high end 30MB/s.