slow gigabit

N

NateD

Weaksauce
Joined
Mar 27, 2010
Messages
66
I have two gigabit enabled machines. They each have different motherboards, but both motherboards use the Realtek RTL8111D NIC. The machines are connected in this way:

machine 1
~15 feet of CAT6
D-Link DGL-4300 gigabit router
3 feet of CAT6
in-wall coupler rated for CAT5e
~50 feet of CAT5e
Rosewill gigabit switch
~15 feet of CAT6
machine 2

File transfers between the machines are slow. :(

I downloaded the PCATTP tool:
http://www.pcausa.com/Utilities/pcattcp.htm
On one machine I run:
pcattcp -R 99999999
One the other I run:
pcattcp -t -f M 192.168.0.4
I see a speed of ~11MB/s (megabytes/second). If I run the second command on the same machine as the first (loopback), then I see 500MB/s.

To try to simplify the test, I replaced the Rosewill switch with a CAT5e inline coupler. I still see only ~11MB/s.

There shouldn't be a problem mixing CAT5e and CAT6, and both can do gigabit.

Do you think a cable or coupler is damaged? Should I get a tester tool? It seems if there was a bad connection or damaged cable that the connection would be reliable. I use the internet through machine 2 every day and it is fine.

Does anyone have any ideas on how to fix my slow transfer speeds? Both machines are running Windows 7 Ultimate, and both have Intel X25-M SSDs. I have set both NICs to 1gbps full duplex. Are there any other settings I should look at? Anything at all that could help?
 
I don't have a good picture of your network in my head, but I will say this.

1) Just because something speaks the 1000Base-T protocol doesn't mean it'll actually run at full gigabit speed. Your D-Link router may fall into this category.

2) The rules for gigabit adapters are different than 10Mb Ethernet or 100Mb Fast Ethernet. You should never hardcode a gigabit adapter to 1000/Full. They will autonegotiate correctly. It's in the spec.

3) Try both machines on the Rosewill switch, not the router.

4) Couplers and gigabit speeds are always a bad idea.
 
Thanks for the reply. :)

1) I'm getting ~11MB/s, which I think is roughly 10/100 speeds right? I see a lot of people on the innertubes complaining about only getting 30-40MB/s on gigabit, which frankly at this point I would be ecstatic to have. :eek: The D-Link router has been very nice in all other respects, I certainly hope it can do better than 11MB/s. If we can prove it can't, I'll definitely replace it.

2) I will put it back to auto negotiate. I'm just grasping at straws at this point.

3) I'll give that a try.

4) I unfortunately don't have a whole lot of choice. The ~50 feet of CAT5e in my network description is all inside the walls. I'm in a 3 story townhome and the wire goes from the 1st floor to the 3rd. It seems to go around some corners, it may be tough to pull a new wire.

Edit: If I set the speed back to auto negotiate, machine 1 shows up with a link speed of 1000, but machine 2 shows up with 100. What could cause this? Both machines have the same NIC and are using the same driver.

I found a Realtek diagnostic tool. The cable test on machine 1 says it is fine, 6m in length, and link 1000. The cable test on machine 2 says it is fine, 32m in length, and link 100.
 
Last edited:
NateD said:
I found a Realtek diagnostic tool. The cable test on machine 1 says it is fine, 6m in length, and link 1000. The cable test on machine 2 says it is fine, 32m in length, and link 100.
Click to expand...

Do you have a real cable tester?
 
10/100 only uses two pairs, but 1000 uses all four. There could be a problem on one of the additional wires in a coupler or cable...
 
I have ordered a cable tester from NewEgg. I'll post how it goes!
 
most new mobos with gigabit ethernet have wire testers in the bios, should check it out if you don't feel like waiting =]

Also, check on the DGL and your OS and see if it shows you what speeds it negotiated at
 
Aha! I had used Realtek's software on machine 2 to check the cables and it said they were fine and 32m. On your recommendation I just ran the BIOS check and got open/open/normal/open, all 0m. I'm pretty sure that means the cabling is borked. Hopefully it is easily identified when the cable tester comes. I am guessing it is the punch down for the in-wall coupler or the jack on the other end, since I did the wiring for those. :) I'll cry if the problem is somewhere in the walls!
 
Cable tester came today. Ha! Turns out it was both the punch down for the in-wall coupler AND the jack on the other end! The punch down is color coded, except that green, green/stripe, orange, orange/stripe are all combinations of green and yellow. I must have guessed at where orange was, and guessed wrong. I also found that in the jack on the other end a wire wasn't touching and that I used a stranded wire jack on a solid cable. I only have jacks for stranded wire, so I went ahead and replaced it with the same and it tests out ok.

Is there any compelling reason to change the jack to one made for solid wire?

Both machines correctly negotiate at 1000 now. Ran my test again... 60.38 MB/s! I know the maximum is 125 MB/s but this is a decent performance for gigabit right? Remember I'm running a network test, so HDDs aren't in play at all.
 
Valnar said:
4) Couplers are always a bad idea.
Click to expand...

T,FTFY :p

Is there any compelling reason to change the jack to one made for solid wire?
Click to expand...
I think solid wire is generally okay in any connector, but the wire itself is more fragile and prone to internal breakage so generally you don't want it for patch cables. Stranded won't work well in punch-down blocks/jacks however, you should always use solid core.

Both machines correctly negotiate at 1000 now. Ran my test again... 60.38 MB/s! I know the maximum is 125 MB/s but this is a decent performance for gigabit right? Remember I'm running a network test, so HDDs aren't in play at all.
Click to expand...
It's reasonable. You probably won't do much better with those NICs. That cheap switch might hurt performance a bit, you could try just connecting the machines directly and see if it is.
 
Guessing that your switch and wiring is fine now, I'd have to blame the Realtek Nic's for the 60mbyte. I have the same chipset in my current fileserver. I max out at 45mbyte/sec on file transfers to windows 7 (with intel gigabit nic) over with samba. I also have another file server made with a server grade intel gigabit nic's and the same hdd setup and I can max out at 120mbyte/s. Why don't I use the sevrer grade? Because it uses double the power... The Realtek's are bottom of the barrel in networking.
 
awesomo is right. If you want killer speeds, you need both machines to have onboard or PCIe Intel NIC's (Server version preferred) and a non-blocking switch. The minimum I'd go is an HP ProCurve 1400-8G. Linksys and Netgear make great 10/100 switches, but they aren't up to snuff on their gigabit offerings.
 
I tried with and without the Rosewill switch and got the same behavior, so that is good, and seems to correlate with everyone's opinion that the Realtek NICs are the bottleneck. The HP ProCurve 1400-8G looks nice and is reasonably priced. I would have got it instead of my $40 Rosewill, but it doesn't seem that I need the upgrade now.

I looked at NICs and found an Intel NICs for ~$34:
http://www.newegg.com/Product/Product.aspx?Item=N82E16833106123
The $53 Intel PILA8470C3 says its a server NIC...
http://www.newegg.com/Product/Product.aspx?Item=N82E16833106109
Otherwise the $87 Intel EXPI9400PT seems to be the choice for a server NIC.
http://www.newegg.com/Product/Product.aspx?Item=N82E16833106010

I'm OK with the ~45MB/s I see for file transfer and ~60MB/s in the network test. If two $34 NICs would double my file transfer, I'd probably do it, but it isn't worth it for me to spend more than that. What performance difference do you think I would see with the $34 NICs, assuming the switch was fine?
 
You want PCIe NICs. If your upgrade is to PCI, it may not be worth it. Get two of these:
http://www.newegg.com/Product/Product.aspx?Item=N82E16833106033

I can't guarantee what performance gain you'll see since it really depends on a lot of things, but I can get a little over 100MB/s with two of these. The server NICs won't offer you anything; the features you pay extra for are not useful in a home environment.
 
You must log in or register to reply here.
Back
Top