slow gigabit, help

macachia · 23 Jan 2006 at 12:12

hi,

i recently changed my network from 100mbit to a gbit lan by puchasing a D-link dgs1005d http://www.dlink.com/products/?pid=229 and used cat6 cables ( 5mtrs ones).

two pcs are attached the switch and am only getting 17-20mb/s, even if i try to transfer big 4gb files.

Both pcs are nforce4 sli and their nvidia network shouldnt be limited to the pci-e bandwidth

any ideas please?

thanks

Skiddley · 23 Jan 2006 at 12:17

Is that MB (Bytes) or Mb (bits) you have quoted?
How are you measuring that?

macachia · 23 Jan 2006 at 12:47

MB (bytes)

Skilldibop · 23 Jan 2006 at 14:12

17MB/s is near 100mbit speed. something somewhere is limiting the throughput to about 100mbit.

Skiddley · 23 Jan 2006 at 14:14

So how are you measuring it?

Skilldibop · 23 Jan 2006 at 14:23

chekeed the link, you got a 10Gbit backplane so it's not that for 4 PCs. What it could be is the switching technique. If it's useing a store and forward strategy for switching then you'll suffer some loss of performance.

V-Spec · 23 Jan 2006 at 15:12

17 Mega Bytes sounds about right for me, the maximum theoretical speed with 100Mbit Fast Ethernet is 12.5 Megabytes per second, (8 bits in a byte, 100Mbit /8 = 12.5M)
17 Megabytes per second will probaly be the limit of your hard disks. as the maximum speed with Gigabit Ethernet is 125 Megabytes per second.

Problem is, with Ethernet and it being a broadcast media, which relies on a rather large MTU, it limits itself to around 75% of its maximum throughput, so in the real world you rarely see an Ethernet link running at its max quoted speed, usually just under.

Skiddley · 23 Jan 2006 at 15:15

I get about 40MB\s to my disk over a gig LAN using SATA 1 disks at both ends. All depends on how you measure it as that determines the number of devices to consider when identifying the bottleneck.

macachia · 23 Jan 2006 at 17:52

how can i measure such rates?

i normally just open an ftp on 1 pc and check the transfer speed

both pcs have sata raptor drives if it matters

benjo plz. · 23 Jan 2006 at 17:55

V-Spec said:
17 Megabytes per second will probaly be the limit of your hard disks. as the maximum speed with Gigabit Ethernet is 125 Megabytes per second.

Yah so when it should be running at 125mb/s its running at 17 mb/s. sounds crap here.

Get the precompile of Iperf from here, http://dast.nlanr.net/Projects/Iperf1.1.1/release.html and run it on each machine, using one as the server and one as the client.

macachia · 23 Jan 2006 at 19:28

results i got:

executing: iperf -s

------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 8.0 KByte (default)
------------------------------------------------------------
[1920] local 10.0.0.1 port 5001 connected with 10.0.0.2 port 3536
[ ID] Interval Transfer Bandwidth
[1920] 0.0-10.0 sec 278 MBytes 222 Mbits/sec

benjo plz. · 23 Jan 2006 at 20:21

sounds like a dodgy cable between the two pcs tbh.

V-Spec · 23 Jan 2006 at 21:27

Phnom_Penh said:
Yah so when it should be running at 125mb/s its running at 17 mb/s. sounds crap here.

Get the precompile of Iperf from here, http://dast.nlanr.net/Projects/Iperf1.1.1/release.html and run it on each machine, using one as the server and one as the client.

He actually said above that it was running at 17MB per second (upper case MB denotes bytes and not bits) As i've already said 17MBps is well above 100Mbit speed and is probably being limited by hdds or something not network oriented.

Toytown · 23 Jan 2006 at 21:52

He actually said above that it was running at 17MB per second (upper case MB denotes bytes and not bits) As i've already said 17MBps is well above 100Mbit speed and is probably being limited by hdds or something not network oriented.

It might well be well above 100Mbit speeds, but it is no where near the speeds of a gigabit lan. Between 2 machines with standard IDE-SATA harddrives he should be looking at about double that figure (the harddrives write speeds will then start to become the limiting factor)

But for 17MB he should definetely be checking out the network cables, the network cards are configured correctly etc etc, this doesnt seem like a limiting factor of anyother hardware, unless he is using slow HD's that can only write at the 17MB a second speed.

benjo plz. · 23 Jan 2006 at 23:05

V-Spec said:
He actually said above that it was running at 17MB per second (upper case MB denotes bytes and not bits) As i've already said 17MBps is well above 100Mbit speed and is probably being limited by hdds or something not network oriented.

MB mb Mb mB don't make much difference, 17MB/s when it should be more like 125MB/s is not normal, however you MB it. Looks like a squashed cable.

17 isn't much about 12.5 tbh, and the guys got raptors, which aren't going to limit the speed that signifcantly.

V-Spec · 24 Jan 2006 at 08:56

Phnom_Penh said:
MB mb Mb mB don't make much difference, 17MB/s when it should be more like 125MB/s is not normal, however you MB it. Looks like a squashed cable.

17 isn't much about 12.5 tbh, and the guys got raptors, which aren't going to limit the speed that signifcantly.

Ok Mb and MB don't make much difference (only a factor of 8

I might add)
Its unlikley that he;d be getting speeds over 100Mbit with a damaged cable, if there was a duplex mismatch then it would be difficult to manage 20Mbit let alone what hes getting now.
If I was him i'd try and get onto the switch and check the interfaces for any errors.

KELTICKNIGHT · 24 Jan 2006 at 14:37

have my wireless bb going straight to gigbit lan card in pc,would l benfit from a router instead of it going straight to lan card

Skilldibop · 24 Jan 2006 at 14:50

Ok Mb and MB don't make much difference (only a factor of 10 I might add

Factor of 8 actually. But TCP hanshaking, Session windowing, frame headers etc. will mess with any exact figures.

Standard Ultra ATA HD should be happy at 400Mb/s or 50MB/s whichever way you prefer. Take a 10% overhead for windows getting in the way and your still talking 45MB/s or 360Mb/s. Given there is a 10Gbit backplane on the switch it's side with a battered cable/ crosstalk/ ***** NIC or some such niggles.

benjo plz. · 24 Jan 2006 at 17:32

V-Spec said:
Ok Mb and MB don't make much difference (only a factor of 8 I might add)
Its unlikley that he;d be getting speeds over 100Mbit with a damaged cable, if there was a duplex mismatch then it would be difficult to manage 20Mbit let alone what hes getting now.
If I was him i'd try and get onto the switch and check the interfaces for any errors.

meh, 17/125 is 13.6% efficiency whatever you want to put after it. If the cables trodden on, it won't conform to Cat5e, which means it won't reach 1000Mbit. So it is perfectly possible to run at 222Mbits-1 with a dodgy cable, you only have to tread on it or bend it or squash it etc to damage, not much at all.
The switch is unmanaged so Its not going to be easy to check for errors. Could be QoS causing problems though.

Skilldibop · 24 Jan 2006 at 17:35

meh, 17/125 is 13.6% efficiency whatever you want to put after it. If the cables trodden on, it won't conform to Cat5e, which means it won't reach 1000Mbit. So it is perfectly possible to run as 222Mbits-1 with a dodgy cable, you only have to tread on it or bend it or squash it etc to damage, not much at all.
The switch is unmanaged so Its not going to be easy to check for errors. Could be QoS causing problems though.

That's true of structured cabling as that's solid core, but patch cables are multicore and are designed to be flexible.