Clients lose network connection at random times !!

[jheyrman]

We have a strange network problem on our LAN.

At certain times (randomly) our network clients seem to lose their network connection to a server, when this occurs they can't connect to one server but are still able to connect to other servers.
This results in many strange things and errors.

"Disk or network error" (MS-Access)
"Cannot save document you must select another file name" (MS-Word en MS-Excell)
"The file is in use by another user… " (MS-Excell)

We have the impression that the problem exists more when network traffic is low.
As a result of this we started pinging the servers every second from a client PC connected to the same switch as the servers. And we discovered that when this error occurs the server in question gives no replies on the ping command. (Sometimes for more than 15 seconds and this happens about 20 times a day)

We discovered now that when we open PcAnywhere sessions to the servers the problem disappears.
The server consoles need to be unlocked and the CPU monitor needs to be visible (to send more data over the PcAnywhere session).
When the sessions are open the servers can respond to the ping command at all times and the clients don't lose their network connections to the servers anymore. When we keep the servers very busy (CPU and LAN) then the clients also seem to keep their connections alive.

Does anyone know of this phenomena or knows a solution?


Here are some specs of the things we have.

Servers:
Poweredge 1300 WinNT 4.0 SP6a (with Intel pro 10/100 LAN card)
Poweredge 4200 WinNT 4.0 SP6a (with 3com 3c980 server LAN card)
Poweredge 2500 WinNT 4.0 SP6a (with Intel 8255xx-based 10/100 LAN card)

Network:
Cisco Ethernet switch model 3524
All servers are connected to this switch and they negotiate 100Mbit Full Duplex.

Things we already tested but didn't help:
We tested with other switches (baystack, 3com) and hubs (Intel)
We changed the cabling.
Forced the auto-negotiation to 100/full 100/half 10/full 10/half
Other network cards (in clients and servers)
Checked out the temp directories on clients (MS KB: Q150943)
We disabled all the power saving on clients (NT4 Server has no power saving)


Best regards
Johan

 

[beerhunter2]

This sounds like a similar network timeout issue that I had on a Windows 2000 server. I hope the same fix works for NT.

Here is how to change the timeout :

The default timeout is 15 minutes.
From a command prompt on the server type
net config server /autodisconnect:-1
for no disconnect.

Caution: Setting this from the command line may turn off auto tuning for the server service. So a better solution is to navigate to

HKLM\SYSTEM\CurrentControlSet\Services\LanmanServer\Parameters

and set the Reg_DWord hex value of "autodisconnect"=ffffffff

If "autodisconnect" doesn't exist, Add it

 

[SoyGalesa]

Ok, we've just had this issue at my workplace and this is how we solved it. Don't let the systems autonegotiate the speed/duplex settings. for some reason, if you allow NT to auto negotiate a speed/duplex setting against a cisco switch it seems to have issues. We found one of our servers had negotiated to half duplex and kept getting knocked off the network by full duplex traffic.

Hard set the speed and duplex on the NIC in the server and also at the port. Also check to see if portfast is enabled on the port, since a server should theoretically never be down, then portfast is not required on any port connecting to a server.

 

[jheyrman]

thanks for the answer SoyGalesa but i've already tested this and didn't solve uot problem....

> Things we already tested but didn't help:
>We tested with other switches (baystack, 3com) and hubs(Intel)
>We changed the cabling.
>Forced the auto-negotiation to 100/full 100/half 10/full 10/half

Can i ask you what network cards you are using ?

Best regards
Johan

 

[dnels]

Johan,

Have you done a 'sho int' on the switchport to see if there are any errors?

On the router connected to the vlan/switch I would do a 'sho ip arp' for the servers ip address to see which mac-address is getting into the arp table of the router. If any.

Then do a 'sho mac-address-table' on the switch and make sure the mac-address associated with the servers ip-address is on the port you expect it to be on.

If the mac-address is not in the mac-address-table on the switch check the error counters on the servers switchport interface.

That's what I'd do to further isolate the problem.

Dave

 

[jheyrman]

Thanks dnels for the answer. I checked the arp & mac tables... Everything seemes to be just fine.

Like i mentioned this does also appear when we put al servers on a simple HUB !

Best regards
Johan

 

[rn4it]

Is both the server and switch/hub configured the same? (ie both set to auto, 100full etc) On the switches, do you see any errors or connections drop?

1 thing I have encountered before, is we had racks that weren't grounded and with some smaller UPS were giving a small amount of electrical interference through the rack to the switch, this caused some weird connection issues.