Web Edition loses connectivity

(Not sure how to get the output of ipconfig into text.) However, both subnets are 255.255.255.0, gateways are 192.168.[0/1].1, no Wins, no IP Routing, appropriate DNS for the gateways.

I've been doing some browsing and it seems it not just me that's having this problem:
http://www.microsoft.com/windowsserver2003/community/newsgroups/dgbrowser/en-us/default.mspx?dg=microsoft.public.inetserver.iis&tid=601cc37c-1b1c-44d1-a320-d8f75109cdd7&cat=&lang=en&cr=US&sloc=en-us&m=1&p=1
http://iis-resources.com/modules/newbb/viewtopic.php?topic_id=3297&viewmode=flat&order=ASC&start=10
http://www.issociate.de/board/post/258461/IIS6_Sites_become_unavailable_until_I_reboot.html

Some people say this is an appropriate solution (I don't see that it's relevant - but then I don't really understand what it's saying!):
http://support.microsoft.com/kb/898060/

Normally, you don't use 2 NICs on Web Edition - however, it's not technically wrong.

Disable Client for MS Networks, File and Print Sharing and NetBIOS over TCP/IP on the public NIC.  Also, uncheck the box for "Register this connection in DNS." Make sure the inside NIC is at the top of the binding order.

Do you need 2 NICs?  They are both private IPs, so why can't you use one?  I think this will simplify things.


That fix is one I've recommended to a number of people.  It seems the hop count increments until it caves in.  You can certainly apply it as it won't break anything else.

Could you explain the rational behind disabling Client for MS, F&P sharing and NetBOIS? Are they just not needed? What will unchecking the Reg in DNS do? How will this help?

I'm sorry to ask more questions. This is a live server and it's our only one at the moment - I won't be popular if I break it!

The two NICs are so we can provide a backup route to this server if the primary ISP connection goes down. We can offer another (temporary) route on another url and port through another ISP. The boss want to have it ready to roll so no-one needs to come to reconfigure and switch cables in the middle of the night. The 192.168.0.* is the orange channel on the primary connection and the 192.168.2.* the orange on the backup connection.

The crop-up period for this problem seems intermittent but it has increased since we put out a couple of other web sites on this server which would increase the "hop count" wouldn't it? So, perhaps it is related after all.

They open a security hole by enabling ports (135, 138, 139, 445, etc) that are not necessary on the external network.  The tighter you make it, the better you will sleep at night.

Removing the ability for the external NIC to register with DNS prevents internal users from attempting to access services from the OUTSIDE NIC - if the services are bound to the internal adapter this will certainly cause problems.

If you want redundant links to 2 different ISPs then you should be looking at something like this:

http://www.xincom.com/products.html

Hop count would increase through normal use, so yes, the more use the faster the problem occurs.

I would look at the twin-WAN routers and remove one NIC - I think you will find things work as you expect them to.

NM

What you suggest on protocols makes sense. I'll have a look into it.

But the change of infrastructure is not an option at this point. We should be moving to co-location (in an ex nuclear bunker) soon which may help the situation. But the machines there would still be dual homed (it's standard isn't it?) to have consecutive IP addresses for hardware redundancy. It's a major concern that we could lose connectivity to machines that are not right next to us. (Especially given the call out for accompanied access to the co-location site is £100 an hour.)

RE the hotfix in http://support.microsoft.com/kb/898060/ : I've looked at the file versions of Tcpip.sys. The hotfix info says I should have a file sized at 333,312 which I have but the versions don't match. It should be 5.2.3790.2453 and I have .1830! That helped. I can't afford to install a patch that might break the machine even more so it's looking like I'm gonna write a script (netsh) to refresh the network card settings every so often.

Just as a follow-up the netsh sript mostly does the trick until we see an increase in activity which shuts down the NIC before the script runs. We refresh every twelve hours and even then people notice the blip in connectivity to their sites so it's not ideal. I have been running mahine patched with 898060 internally without any probems and so have applied it to our new co-located servers which should be live soon. Here's hoping I get some joy. I'll keep you posted!