How fast should gigabit ethernet be?

[tsigo]

I'm upgrading the connection between my main workstation and my fileserver to be gigabit ethernet, since those are the two computers that I use the most.

I got two Intel 1000/MT Pro cards and a Linksys 8-port gigabit switch (http://www.linksys.com/products/product.asp?grid=35&scid=42&prid=529). Transfer speed between them maxes out at about 15,000KBps, and it's not even a constant speed. One cable is Cat 6 and the other is Cat 5e. Both drives are 7200 RPM. Both are in 1000baseT/Full mode.

What kind of speeds should I be getting? What can I check or change to get the full speed?

[tim_m]

what kind of hard drives? ide? the network is not your bottlenexk rather your computers and their components (hard drives/pci bus).

to get the fastest speed you need freakin fast scsi and may as well get a mobo with 64 bit pci slots and scsi controller cards

you see it's a combination of things, it will only operate as fast as the slowest component.

[tsigo]

Yeah I figured it was mostly the hard drives. Thanks.

[drozenski]

well heres the Rule *note this is in a perfect world only*

10Base-T = max transfer of 1Mbps
100Base-T = max transfer of 10Mbps
1000Base-T = max transfer of 100Mbps

pretty simple

[randyc]

how many Kilobytes is a Megabit?

[Cactus Jack]

Originally posted by drozenski
well heres the Rule *note this is in a perfect world only*

10Base-T = max transfer of 1Mbps
100Base-T = max transfer of 10Mbps
1000Base-T = max transfer of 100Mbps

pretty simple

Where did you get these numbers? Those aren't even close. 100mbit on my system is around 7-8mbyte/s. And depending on your gigabit card, motherboard, hard drives, etc. you can get speeds pretty close to 1 gigabit. Even with a decent system, 200-300 megabit is within reach.

[skritch]

Originally posted by drozenski
well heres the Rule *note this is in a perfect world only*

10Base-T = max transfer of 1Mbps
100Base-T = max transfer of 10Mbps
1000Base-T = max transfer of 100Mbps

pretty simple

These are off by an order of magnitude.

[tim_m]

bit to byte is simple, just divide by 8 (right?)
10 mbit => 1.25 mbyte
100 => 12.5
1000 => 125

that's the theoretical conversion factor (to be more precise) but it never reaches the max due to other limitations (why it only does maybe 7-8 mbytes, Cactus Jack)

you would need a pretty good setup to get a full 125 mbytes/sec sustained transfer

for randyc...
1 mbit / 8 bytes per bit = 0.125 mbyte * 1000 k per m = 125 kbytes

[skritch]

Originally posted by tim
bit to byte is simple, just divide by 8 (right?)
10 mbit => 1.25 mbyte
100 => 12.5
1000 => 125



Yes, but the person I quoted had 10^t=1Mbps (megabit per second), not 1 MBps (megabyte per second), and so on down the line.

The former's low by an order of magnitude, the latter's a good, easy-to-remember metric.

[Nate7311]

Most likely It's the 32bit PCI bus. The extra bandwidth of a 64bit and/or 66mhz slot make a considerable improvement.

32bit:
http://www.digit-life.com/articles2/gigeth32bit/gig-eth-32bit-2.html

Versus,


64bit:
http://www.digit-life.com/articles2/gig-eth-64bit-amd760/

[Zardoz]

It does seem low to me. I get 8-10MB/s (Megabytes) on my 100Mb network. 15MB/s does not seem that much better. I would expect at least 50MB/s or better, but some of the points made so far seem very valid as to why it is not...

[tim_m]

Originally posted by skritch
Yes, but the person I quoted had 10^t=1Mbps (megabit per second), not 1 MBps (megabyte per second), and so on down the line.

The former's low by an order of magnitude, the latter's a good, easy-to-remember metric. i was never quite sure about which was what letter case and since most of the time one just types in lower case, you can't tell. i know technically that lower case m is milli as in millimeters and capital M is mega like megawatts but i wasn't totally sure about b/B. thanks for clearing it up

[Phantum]

You'll never see theoretical bandwidth. Not even under the best conditions, never. So it's always practical to take the max and subtract. Generally on 100Mbps you see about 3MBps. Even if you flood it, even with Raptors, hell even with SCSI you'd maybe see 5MBps, but that's it.

[Boscoh]

Interesting....I routinely get about 10MB/sec on 100Mb host-host file transfers.

[Phantum]

Originally posted by Boscoh
Interesting....I routinely get about 10MB/sec on 100Mb host-host file transfers.

Riiiiiight, are you sure it's 10MBps? That seems a little high.

[DeepFreeze]

around 7-8MB /sec

[Boscoh]

I'm sure...I'll see if I can get some screen cap's for you when I have some time if you really want to see it.

[Phantum]

Originally posted by Boscoh
I'm sure...I'll see if I can get some screen cap's for you when I have some time if you really want to see it.


OOOOOH yeah pics gimme.

[seanx]

Originally posted by Phantum
OOOOOH yeah pics gimme.

On my 100mbps network I get close to 10MB/s also usally at least 9.

[ambit]

Originally posted by seanx
On my 100mbps network I get close to 10MB/s also usally at least 9.
i'll agree to that as well

[[]wave]

Originally posted by Phantum
Riiiiiight, are you sure it's 10MBps? That seems a little high.
what? :confused: :confused: :confused:

[Laforge]

Originally posted by drozenski
well heres the Rule *note this is in a perfect world only*

10Base-T = max transfer of 1Mbps
100Base-T = max transfer of 10Mbps
1000Base-T = max transfer of 100Mbps

pretty simple

Huh?

1000Base t is limited to 100Megabits per second? No..that's 100Base-t


Linky (http://www.cisco.com/warp/public/cc/techno/media/lan/gig/tech/gigbt_tc.htm)

[mogman42]

on my old 100baseT network I could get ~10-11MB/sec and with two Intel Pro/1000MT nics and an SMC gigabit switch I get 30MB/sec btw two IDE systems.....

[Boscoh]

I think he meant MBps instead of Mbps....

Even then, that isnt correct. I've seen bursts of traffic on 1000BaseT of well over 100MBps, the most I've ever seen on a 100BaseT network was just shy of 11.5MBps.

I'd say that list is what you should expect in ANY 1000BaseT, 100BaseT or 10BaseT network, not just in a perfect world. In a perfect world, you'd want as close to the REAL theoretical max as possible. And certainly more than any of those numbers, especially dealing with 1000BaseT.

[adc1986]

Keep in mind that most network cards are cheap $10 junk.

Unless you want to spend 30-40 even for a decent 100mbit card you will usually only see 60-75% of what the network card actually says it will do. Same way for the gigabit.

If you buy a cheaper 1000baset card, you can't expect to see performance gains, but you should see atleast 25megs/sec. The server cards are the ones you need to look at if you want some great transfer rates.

Unless you want to shell out the cash for those and have a 64bit pci bus, you will have to live with the slow transfers.

[str8eiht]

>on my old 100baseT network I could get ~10-11MB/sec and >with two Intel Pro/1000MT nics and an SMC gigabit switch I get >30MB/sec btw two IDE systems.....

Same here, i did a tonne of benchmarking / config changes when i first got my Gig cards, i have connected via x-over though, no switch yet. I get about 30-35 mb/sec, just below my sustained transfer speed for the HDDz

[Zardoz]

Here is a FTP transfer on my 100Mb network. This file is just over 1GB in size. I ran this test 3 times with the same results. This is a screen shot of the transfer in mid coerce. Speed was always above 11MB/s (Megabytes) the ftp server is a 900 MHz Athlon with 256MB of RAM. The NIC in the server is a 3Com 3c905cx and uses 3 120GB IDE HDs in a software RAID 5. The desktop system is a Shuttle with a 1900+ Athlon using the onboard SIS NIC and IDE hard drive as well.

http://www.laggedout.com/zardoz/image/networking/networkspeed.jpg

[TrueBuckeye]

One thing to keep in mind is that most network protocols have an overhead which will eat into your max bandwidth. The fastest protocol is NetBEUI with a normal loss of 2-5%, then you have IPX, and finally TCP/IP which can cost you 10-12%.

So, 100mpbs network is 12.5MB/s max. Running TCP/IP you would expect to get around 10MB/s on an excellent connection. For gigabit, you are looking at 100MB/s max, assuming no bottlenecks.

Also keep in mind that a pci bus runs 32bits per cycle and 33,000 cycles per second, so the max throughput of the bus is 1.056Gbps. So running a gigabit pci nic on a standard desktop the network traffic can completely saturate the pci bus. So any other systems using the same bus (sound cards, etc) will lower your max network speed.

This is why servers run 64bit pci bus and why new standards are coming out like pci-x which will double the speed of the bus, giving you a max throughput of 4.224Gbps. You toss 2-4 gigabit cards in there with teaming and NOW you've got some serious bandwidth. :D

[PopeKevinI]

Originally posted by Phantum
Riiiiiight, are you sure it's 10MBps? That seems a little high.

That sounds about right, with a good pair of NICs and (if necessary) a fast switch.

On a 1Gb test setup, we got 99 MBps using two dual Xeon systems with big 2Gb Fibre Channel RAIDs. Intel Pro/1000XT cards, I believe.

A slow switch or NIC can slow it down considerably.

[Snugglebear]

Originally posted by PopeKevinI
That sounds about right, with a good pair of NICs and (if necessary) a fast switch.

On a 1Gb test setup, we got 99 MBps using two dual Xeon systems with big 2Gb Fibre Channel RAIDs. Intel Pro/1000XT cards, I believe.

A slow switch or NIC can slow it down considerably.

Don't forget to toss in a good OS and application to transfer that data with. I've still not gotten close to the 11-11.5 MB/s that I had running Samba 2.x on Solaris 8. And that was with a P2-266, a Celeron 466 on Win2k, Intel nics both, and a linksys switch.

[PopeKevinI]

Originally posted by Snugglebear
Don't forget to toss in a good OS and application to transfer that data with. I've still not gotten close to the 11-11.5 MB/s that I had running Samba 2.x on Solaris 8. And that was with a P2-266, a Celeron 466 on Win2k, Intel nics both, and a linksys switch.

Well, your switch isn't exactly a speed demon.

Windows 2000 had no problems whatsoever on the hardware we were using. 2003 was marginally faster (.1 MBps I think), probably due to the apparent increase in hard drive read speed.

[TrueBuckeye]

Originally posted by PopeKevinI
Well, your switch isn't exactly a speed demon.

Windows 2000 had no problems whatsoever on the hardware we were using. 2003 was marginally faster (.1 MBps I think), probably due to the apparent increase in hard drive read speed.

From what I've heard, it is probably more likely due to the improved TCP/IP stack in XP over 2k. I heard the code was tightened up quite a bit and in excellent conditions you could see a 2-3% speed increase.