Slow Gigabit connection

[jsteph]

I am experiencing very slow network traffic, and the cables have been swapped between a known-good cable on another machine on the rack.

It's a BroadCom NetXTreme Gigabit onboard adapter, running at 100mbps; we haven't upgrated the routers yet. The machine is a Dell Poweredge, Quad PIII processor with 2 Gig Ram, running Win2k Advanced Server. There is nothing and nobody else on this box--the rest of the company doesn't know it exists yet, it's a development machine.

I copy a 10 meg file from any workstation, and it takes about 25 seconds to this machine. My simple math tells me that's about 400K per second, or 3.2 mbps. Another known-good machine gets the file in 5 seconds, or 16mbps. I swap the cables, so the 'bad' machine is getting the file via the same physical route as the good one, and it gets the same bad peformance--so the problem is somewhere inside the machine.

I looked at bcastner's link in another similar thread that talks about some regisry settings. The first setting--....\Services\e100bx does not exist, nothing prefixed e100 is there (maybe because it's gigabit, but no e1000 either). But I tried the other one--the TcpWindowSize and that seemed to slow it down even more.

Can anyone point me to another setting that I might try?
Thanks,
jsteph

 

[franklin97355]

There has been comments that the onboard nics are not all that good on some servers. We sped ours up by adding regular nic cards, helped a lot.

 

[jsteph]

Yes, I've heard that too. But that might explain at most a 10% or 20% difference in speed. This is 400% difference. And the 'known-good' machine has an onboard Nic also.

What I'm leaning towards is some registry setting, or, I'm wondering if, with gigabit being relatively new, could there be any known issues with older routers (ours are over 5 years old), that don't play well with a gigabit nic, even though the gigabit is self-throttled to 100mbs?

It's a huge difference, so I was hoping for one of those 'light bulb' moments where someone had the same thing and it was some obvious setting that got hosed or something.
--j

 

[jimbopalmer]

Duplex is Auto, right?

I tried to remain child-like, all I acheived was childish.

 

[jsteph]

Yes, that's auto. There are two onboard gigabit nics on this, the other is disabled, so I don't think there's any conflict there.

In the Advanced properites of the nic, I have:

802.1p QOS - Disabled
Checksum Offload - Tx Tcpip Checksum
Ethernet@Wirespeed - Enable
Flow Control - Disable
Jumbo MTU - 1500
Speed & Duplex - Auto
Wakeup Capabilities - Both
WOL Speed - AUTO

...does anything look fishy there?
Thanks,
--Jsteph

 

[TimRegester]

Change the duplex and speed settings to 100BASE-T Full duplex at both the server and switch side. If this fixes the issue it will be, as is quite common, an autonegotiation issue between the two devices. This would explain slow throughput adequately.

If this changes nothing you can switch them back, but for servers it is always good practice to fix the speed and duplex.

 

[jsteph]

Tim,
Thank you. What we ended up doing is installing new drivers and enabling both nics. I'll have to go back and see if just setting both the switch and the nic to fixed will do it also. But it seemed that having two nics on board somehow necessitated enabling both and doing a round-robin with the ip's.
---jsteph