In reply to a post by joconnell:

router to ... support jumbo frames on the LAN interface

Well the router is a Thomson TG585 v7 with 100Mbps ethernet ports, so jumbo frames don't apply do they? The Netgear GS605 switch (into which the router is plugged in order to share the internet connection) has no config options and the gigabit NICs have jumbo frames disabled (because there are also 100Mbps devices on the network)

Yes for your setup you would probably only use jumbo frames temporarily as part of testing transfers. There are consumer routers with gigabit interfaces so I left that in for the more general case.

I had looked into jumbo frames for a work situ with a similar result that it's not yet convenient to run the jumbo and non-jumbo (or 100M) stuff on separate networks (or VLANs). The non-jumbo stuff included things like shared printers and even some Mac laptops which I found a surprising omission.

I'm guessing a mixed setup might work but with more retransmissions so not necessarily helping.

In reply to a post by joconnell:

Next step, try direct connection between PCs.

Okay, I've connected both PCs directly with a 2m Cat5e cable, assigned static IP addresses and run speed tests using LAN Speed Test, and I get speeds of around 140Mbps.
Using Ixia QTest reports 200Mbps. With 8KB jumbo frames enabled on both NICs, that result is doubled, but I've left jumbo frames off as the network is a mix of gigabit and 100Mbps ethernet.
Running iPerf reports speeds of 9.5Mbps!
A 2 gigabyte file copy from the slower PC to the faster PC takes 1 min 45 secs - I've run the test in that direction as I've reasoned that reading a file from a slow machine and writing to a fast machine reduces skewed results resulting from a slower system.

Now given that the network consists of one 2m patch cable (not a crossover cable) linking 2 gigabit NICs, I'm at a loss to explain why performance is so poor. Others have reported speeds of 700Mbps or more on gigabit networks. How are others testing their networks?

The cable is brand new and I've tried a few cables, no difference to the test results. Unless one or both of the NICs cannot perform to gigabit speeds due to a fault or just shortcomings of the NIC?

Has anyone any more suggestions?

Edited by joconnell (Thu 02-Jun-11 00:46:10)

I take it you mean 90-100 megabytes per sec...

I've just copied two video files acroos my lan. The first file was 3.83GB & took 53 seconds. The second was 3.95Gb & took 47 secs.

Snapshot 1
Snapshot 2

Both pc's have Gigabyte motherboards with onboard gigabit nics & both are running windows 7.

Edited by deleted (Thu 02-Jun-11 08:19:33)

Running iPerf reports speeds of 9.5Mbps!
A 2 gigabyte file copy from the slower PC to the faster PC takes 1 min 45 secs - I've run the test in that direction as I've reasoned that reading a file from a slow machine and writing to a fast machine reduces skewed results resulting from a slower system.

Thats similar to speeds I was getting with XP & ide drives. I couldn't pinpoint the problem but when I installed Win 7, speeds immediately improved to 50Mbs. This is on an old Abit socket 754 with onboard gb nic.

Quad core Gigabyte to Abit socket 754

In reply to a post by Als:

That's similar to speeds I was getting with XP & ide drives. I couldn't pinpoint the problem but when I installed Win 7, speeds immediately improved to 50Mbs. This is on an old Abit socket 754 with onboard gb nic.

Quad core Gigabyte to Abit socket 754

Do you have jumbo frames enabled for your gigabit nics and is every device on your LAN gigabit?

So was Win XP the weak link? Do you have a win7 PC with IDE drives which now performs much better? What's the spec of the machine you were copying from/to?

Edited by joconnell (Thu 02-Jun-11 12:15:20)

Odd, isn't it. Tests measure what tests measure, and may or may not reflect the ultimate capability of two NICs in two computers.

If you can find some relevant Linux tests you might like to cut out the Windows IP stack and see if that sheds any more light.

"A 2 gigabyte file copy from the slower PC to the faster PC takes 1 min 45 secs - I've run the test in that direction as I've reasoned that reading a file from a slow machine and writing to a fast machine reduces skewed results resulting from a slower system."

Rather than assuming, you might want to actually try it, if it's not going to be difficult.

If the slow computer is doing the reading, surely that WILL limit the transfer speed as there is nothing to send if there's no data available? If the slow computer is doing the writing, it's not so much a bottleneck as data it receives can in principle be cached in memory before being written to disk? Not saying that's 100% what happens, but I believe NTFS on Windows does by default write-behind like that, because it does provide some apparent performance gain when writing. The downside of write-behind like that is that data in memory but not on disk is vulnerable to being lost if Windows crashes or if power fails, etc.

No, I don't have jumbo frames enabled as we have a couple of laptops with 100Mb cards.

I'm quite sure XP was part of the problem. I could get good rates between two Win7 machines but not from Win7 to XP. The Win7 machines have sata drives whereas the XP machine was all IDE. As soon as I installed Win7 on the old XP box the nw speed improved by at least 5x.

The old XP, now Win7, box is Abit NF8, 3000 Athlon, onboard Realtek gb nic,1G ram, 3 IDE hard drives, 1 IDE dvd.
The other Win7 boxes are Gigabyte, onboard Realtek gb nics, Intel cpus 4 & 6GB rams, all sata drives.

When I had XP on the Gigabyte boxes I got reasonably good lan speeds but my GS608 was playing around on a couple of ports. Replacing it with the Zyxel switch cured the slow speeds. But I still had the slow speeds to the Abit NF8 Athlon box which was also on XP. After trying everything I could think of I decided to try Win7. There was an immediate improvement. Its not perfect & I'm guessing that its either the pci bus or ide drives that are the limiting factor.

My daughter's bf checked my ethernet cables with very expensive pro test gear & found nothing wrong.

I've got another old XP box I could fire up & do some tests. Its a cheap & nasty Elite or ECS board, iirc. I have a netgear 311 gb nic that I can try oout & see what speeds I get.

In reply to a post by Als:

No, I don't have jumbo frames enabled as we have a couple of laptops with 100Mb cards.

I'm quite sure XP was part of the problem. I could get good rates between two Win7 machines but not from Win7 to XP. The Win7 machines have sata drives whereas the XP machine was all IDE. As soon as I installed Win7 on the old XP box the nw speed improved by at least 5x.

The old XP, now Win7, box is Abit NF8, 3000 Athlon, onboard Realtek gb nic,1G ram, 3 IDE hard drives, 1 IDE dvd.
The other Win7 boxes are Gigabyte, onboard Realtek gb nics, Intel cpus 4 & 6GB rams, all sata drives.

When I had XP on the Gigabyte boxes I got reasonably good lan speeds but my GS608 was playing around on a couple of ports. Replacing it with the Zyxel switch cured the slow speeds. But I still had the slow speeds to the Abit NF8 Athlon box which was also on XP. After trying everything I could think of I decided to try Win7. There was an immediate improvement. Its not perfect & I'm guessing that its either the pci bus or ide drives that are the limiting factor.

My daughter's bf checked my ethernet cables with very expensive pro test gear & found nothing wrong.

I've got another old XP box I could fire up & do some tests. Its a cheap & nasty Elite or ECS board, iirc. I have a netgear 311 gb nic that I can try oout & see what speeds I get.

Thanks for all your info.

The only things I can think is that I should use a crossover cable rather than a patch cable to connect the two PCs, though with gigabit NICs I've been told a crossover cable isn't needed. I think I saw a van from a datacomms firm parked at a house near me, I might have a word and see if I can get my home network checked out, at least that way I'll know it's alright, which is the point of the testing I'm doing now.

But ultimately, the best bet is to test between a couple of modern Win 7 PCs, or as someone suggsted, use Linux to test.

In reply to a post by Anonymous:

Odd, isn't it. Tests measure what tests measure, and may or may not reflect the ultimate capability of two NICs in two computers.

If you can find some relevant Linux tests you might like to cut out the Windows IP stack and see if that sheds any more light.

That's a good idea, I wonder if I could run tests off a CD booted version of Ubuntu or whatever? Can Linux access the local windows disks?

I don't think I'm going to get much further with my testing until I can get another up to date Win7 PC with a gigabit NIC. Here's my current results of testing by copying two files from one 10 year old PC to an 18 month old PC, both with gigabit NICs.

Text

1 23 45 67 89 1011

direct connect with patch cable 2,090,110,986 in 1.35 mins [168 Mbps]1,482,419,895 in 1.09 mins [164 Mbps] ------------------------------------------------------------------------------------------connected via switch with router (100Mbps) plugged into switch 2,090,110,986 in 1.40 mins [160 Mbps]1,482,419,895 in 1.12 mins [158 Mbps] ------------------------------------------------------------------------------------------connected via switch with no router plugged in 2,090,110,986 in 1.40 mins [160 Mbps] 1,482,419,895 in 1.09 mins [164 Mbps]

At this point the best I can do is to establish whether the tested speeds on my home gigabit LAN are close to the above results, that way I'll know that at least the LAN cabling is terminated okay at the RJ45 modules and at the patch panel.

Thanks for all your help so far.