Any LAN experts? Not a broadband question!

[Inkblot]

Nothing to do with IDNet but a problem I am having at the moment I'll try and keep it simple but keep all the information in...

Client had a single server (Win2003r2) with a number of PC's all connected to a simple 10/100 switched network, 3 switches in all, not stacking, just using uplink beween them. Browsing the server was fine and ping times from any workstation to the server were perfect regardless of which switch you were connected to:

Pinging server1 [192.168.1.5] with 32 bytes of data:

Reply from 192.168.1.5: bytes=32 time<1ms TTL=128
Reply from 192.168.1.5: bytes=32 time<1ms TTL=128
Reply from 192.168.1.5: bytes=32 time<1ms TTL=128
Reply from 192.168.1.5: bytes=32 time<1ms TTL=128

Ping statistics for 192.168.1.5:
   Packets: Sent = 4, Received = 4, Lost = 0 (0% loss)
Approximate round trip times in milli-seconds:
   Minimum = 0ms, Maximum = 0ms, Average = 0ms

About 3 weeks ago we added a 2nd server (Win2008r2) to the domain and replaced the old 10/100 switches with Gigabit ones, this time stacking ones. The AD Schema was updated before adding the 2nd server.

Browsing or pinging the original server from anywhere on the network is exactly as it was - perfect. The 2nd server however is slow to browse and pinging it gives me strange results:

Pinging server2 [192.168.1.7] with 32 bytes of data:

Reply from 192.168.1.7: bytes=32 time<1ms TTL=128
Reply from 192.168.1.7: bytes=32 time=523ms TTL=128
Reply from 192.168.1.7: bytes=32 time<1ms TTL=128
Reply from 192.168.1.7: bytes=32 time=520ms TTL=128
Reply from 192.168.1.7: bytes=32 time<1ms TTL=128
Reply from 192.168.1.7: bytes=32 time=517ms TTL=128
Reply from 192.168.1.7: bytes=32 time<1ms TTL=128
Reply from 192.168.1.7: bytes=32 time=512ms TTL=128
Reply from 192.168.1.7: bytes=32 time<1ms TTL=128
Reply from 192.168.1.7: bytes=32 time=526ms TTL=128

Ping statistics for 192.168.1.7:
   Packets: Sent = 10, Received = 10, Lost = 0 (0% loss),
Approximate round trip times in milli-seconds:
   Minimum = 0ms, Maximum = 526ms, Average = 259ms

As you can see, the ping alternates between normal and slow. What's worse is that the slow pings are sometimes ~700ms, sometimes ~300ms or sometimes not there at all - I always get 100% packets returned, sometimes all at 1ms instead of the fast/slow alternating as shown above.

Pinging out from the 2nd server to the 1st is the same. alternating between slow and fast.

Anything involving the 1st server is fine, anything involving the 2nd server is slow...but exactly how slow varies, sometimes even running 2 pings simultaneously from the same machine will have different results.

More info:
Both servers are connected to the same physical switch
Re-patching 2nd server into a different port makes no difference (Still fast/slow)
Patching another PC into the port previously occupied by 2nd server makes no difference (Still fast to server1, slow/fast to server2)
Changing patch leads makes no difference
No IP filtering is enabled
Firewall configuration has been checked
No filtering is taking place at the switch
DCDIAG /Test:DNS reports no errors

Based on the testing I have done I'm happy that it's not related to the switches or the workstations but I have never seen this on a server before.

Anybody seen this before or have any ideas that I have missed?

Thanks!

[Rik]

Sorry, Inky, I can't think of anything - but it's weird. Can you try changing the IP address of the server?

[Inkblot]

Thanks Rik, believe it or not I may have just found the answer right after posting my cry for help! Server1 has 2 NICs, NIC1 in use and NIC2 disabled. Server2 also has 2 NICs, NIC1 in use but in this case NIC2 although not connected to anything (No patch lead in it) was not disabled. It was set to pick up an IP via DHCP but was of course failing and autoconfigring itself onto a 169.254.x.x address. There is no teaming of the NICs, no bonding, no nothing at all but somehow disabling NIC2 has (So far) meant my ping times are now perfect, more testing will need to follow but I've never had ping times as consistent from this server before so fingers crossed (Can't find the icon for it) I have somehow stumbled onto the answer

[Rik]

  You mean, Inky? One for the memory bank, that. I must admit I disable the second port on my machines, but that's just to stop them consuming resources for something I'm not using.

[Tacitus]

You mean, Inky? One for the memory bank, that. I must admit I disable the second port on my machines, but that's just to stop them consuming resources for something I'm not using.

Interesting.  My PowerMac has two NICs only one of which is in use.  Not having had any problems I've never thought to disable it, but I have now just in case.   

[Rik]

Just remember which socket is the live one, Tac.

[Inkblot]

Interesting.  My PowerMac has two NICs only one of which is in use.  Not having had any problems I've never thought to disable it, but I have now just in case.   

Similarly, my desktop PC has 2 and I'm certain that I've had then both active with only 1 connected before now with no problems. Usually I do disable the 2nd NIC on a server if I don't need it but this was part of a bigger job with not much time allowed and it was overloooked, I didn't think to check it until today as I've never seen that before but over the weekend I'll test it a little more thoroughly.

[Rik]

It is weird, Inky, they should be completely independent. Might be worth a look into the BIOS to see if there's anything odd about the setup.

[esh]

Some of the Intel multiple-socket NICs run on a single chip, I believe. Hence they are not completely independent. But I would wager it's more likely a driver issue doing something silly.

[Dangerjunkie]

Hi,

I'm not sure if you're using a managed or unmanaged switch. Depending on the make of the switch (Ciscos being particularly blame-worthy) I've found that it can be beneficial to hard-set the speed and duplex of the ports and machines rather than letting it autonegotiate.

We've had particular problems with Dell PCs on Cisco switches where both are 1000 full-duplex capable but one autonegotiates from high to low and the other from low to high and they end up settling on 100 half-duplex. Another issue has seen some links where one end ends up set to 1000 full and the other to 1000 half. The results of both of these scenarios is terrible performance of the kind you are seeing.

If you're running a managed switch with VLANs a duplex mismatch can affect a whole VLAN and ripple through every switch in the network. We had one where this happened, somebody plugged an unauthorised piece of kit in on the 1st floor switch and the effect turned the whole VLAN to half duplex and slowed the 30 Mbps Internet connection on the basement switch down to 1mbps half duplex!

Cheers,
Paul.

[Inkblot]

Thanks for all the help, much appreciated! Cisco would be nice but budgets dictate something cheaper. Currently, everything is set to auto-negotiate but seems to be settling with a 1gbps connection (Not sure about duplex, not on site at the moment). I've not yet had the chance to go into the BIOS but will do so next time I am at that site and the client can spare me the machine for a few minutes - might be a little while before I get the chance but I will try and remember to do it at some point.

I ran some simple ping tests over the weekend and it does indeed appear that the 2nd NIC was the problem as with it disabled my ping times are consistently <1ms and the users can browse the network without any issues.

[Rik]

I do like a happy ending.

[Dangerjunkie]

Just out of interest. If you have the 2nd NIC enabled do both NICs have a default gateway if you open a command prompt and type ipconfig /all

Cheers,
Paul.