Something stupid my boss did at work

[a_unique_person]

I work in a reasonably dysfunctional IT environment. We merged two sites a few years ago, and instead of merging the subnets, kept them separate, to save a few dollars. A router was used to connect the two subnets, and send them out to the WAN.

There is now a new manager, who wanted to merge the two subnets. So what has happened is that the switches that make up the two subnets have been directly joined to each other. I don't know a lot about networks, but IIRC, that's not the right way to do it. We seem to be having various weird network issues happening now.

So, can you just leave the switches as is, or do we try to work out how it was connected before it was changed and go back to the old configuration?

[lionking]

I work in a reasonably dysfunctional IT environment. We merged two sites a few years ago, and instead of merging the subnets, kept them separate, to save a few dollars. A router was used to connect the two subnets, and send them out to the WAN.

There is now a new manager, who wanted to merge the two subnets. So what has happened is that the switches that make up the two subnets have been directly joined to each other. I don't know a lot about networks, but IIRC, that's not the right way to do it. We seem to be having various weird network issues happening now.

So, can you just leave the switches as is, or do we try to work out how it was connected before it was changed and go back to the old configuration?
Is there any other?

[huw-l]

there isn't really enough information to work out how bad an idea this was, but it was still a bad idea. One has to assume the router was doing something.

Directly connecting two separate subnets may sort of work but it won't win you any networking best practice awards.

[ddt]

I work in a reasonably dysfunctional IT environment. We merged two sites a few years ago, and instead of merging the subnets, kept them separate, to save a few dollars. A router was used to connect the two subnets, and send them out to the WAN.

There is now a new manager, who wanted to merge the two subnets. So what has happened is that the switches that make up the two subnets have been directly joined to each other. I don't know a lot about networks, but IIRC, that's not the right way to do it. We seem to be having various weird network issues happening now.

So, can you just leave the switches as is, or do we try to work out how it was connected before it was changed and go back to the old configuration?

It probably is a very bad idea. I'm not exactly clear on your configuration, so I make one up myself for showing the point.

Site1 uses IP addresses of the form 192.168.1.x
Site2 uses IP addresses of the form 192.168.2.x
(those addresses are all only for private use and should not get loose on the internet).

Each computer - and router - has a routing table. That routing table says for each IP address whether it is directly reachable (i.e., not via a router) or not.
For directly reachable addresses, it also says over which network interface (most computers have only one, but routers typically have more).
For not directly reachable addresses, it defines the gateway, i.e., the router it has to send the IP packets to that will relay it further. Of course, that gateway has to be a directly reachable address.

Now imagine in the above situation, that Site1 and Site2 are linked with a router. The router has three interfaces: one to Site1 (with IP address 192.168.1.254), one to Site2 (with IP address 192.168.2.254), and one to the outside. The first two interfaces are physically linked with the respective switches - but a switch knows nothing about IP addresses.

The routing table of PC1 at Site1 is simple: addresses of the form 192.168.1.x can be reached directly, and all others have to be sent to 192.168.1.254 for relaying - that includes the addresses of Site2.

The computers of Site1 are linked together - and connected to the router - with a switch, but the switch knows nothing about IP addresses.

Now imagine you pull away the router from this scenario. All outbound traffic from Site1 - and that includes traffic to Site2 - first heads for the router, but the router is not there anymore, so it can't reach the first step in its relaying process, let alone its end destination.

From your story, it seems it still sometimes works, so a bit more has happened, but it certainly is a bad idea to do this if you don't know the basics of IP networking.

[Cainkane1]

The worlds full of stupid managers. When I was a young man my manager told to close our account with the post office as we were moving and were going to use another nearer office. I turned in the key to the PO box and when I got back my manager asked the stupidiest question I've ever heard come out of a womans mouth. She asked "why did you turn in the po box key"? I mean anyone should know that when you close the account you have to turn in the key.

[fxm]

The idea wasn't in and of itself a bad one, it just may have been badly carried out. It depends on how the network is configured, what kind of switches you have, and how the change was carried out.

Some switches can operate at layer 3 (i.e., they are also routers). If the two subnets were on separate VLANs, the switches could be configured to route directly between the two VLANs rather than via the WAN router.

If the switches aren't layer 3 capable, but the two subnets were adjacent to each other, they could have been merged into one larger subnet.

In either case, though, the hosts on the network would have to learn about the network configuration change somehow. If the hosts are statically configured, every host's configuration would have to be changed manually. If the network is set up to distribute network information from a central server (via DHCP, for example), then the hosts would need to be rebooted or otherwise kicked to go and update their network configurations from the server.

Simply reverting the network connections back to the way they were would not be advisable, as it could actually cause more problems. You really have to find out how the switches, routers, and hosts were configured previously, what changes were actually implemented, and then track down any errors that might have been made during the change.

[Kopji]

Might start by looking at the dhcp server configuration. If each network had an individual dhcp server, one might need to be turned off now that they are a single network. Could try setting the IP lease interval to something short like a few hours instead of days. The dhcp servers probably also had a gateway value which is no needed. Any static device might need to be remapped.

If the subnet masks of the two networks were different...

[a_unique_person]

Might start by looking at the dhcp server configuration. If each network had an individual dhcp server, one might need to be turned off now that they are a single network. Could try setting the IP lease interval to something short like a few hours instead of days. The dhcp servers probably also had a gateway value which is no needed. Any static device might need to be remapped.

If the subnet masks of the two networks were different...

I was thinking about the DHCP, yes, there are two.

The masks are

255.255.255.0 on one

and

255.255.254.0 on the other.

[Kopji]

Since you used to have two separate networks and now have one that is connected via a switch rather than a router, I think the dhcp server is the area to look for a problem.

Try any advice at your own risk, but I would try stopping the dhcp server process on one of the servers and see if your problems gradually go away. The one with a mask of 255.255.255.0 is a good one to keep.

The dhcp has three values it gives out:

1: An IP address. This is leased on a time schedule that you program into the server, but is usually from 1 to 3 days. Typically, at 50% of the lease cycle a renewal will be requested. If the PC already has a good IP number, it is just renewed. If not, a new number is allocated from the pool.

If you have two servers handing out numbers from overlapping pool ranges, one server will think a number is available when it is not. Also, if you have static mapped values within the pool, that is bad too.

2: The dhcp hands out the subnet mask value. If you change this, it is given out when the lease is renewed.

3: the gateway number is usually the router on your kind of network.

The servers themselves will have a static IP mapping, the mapping needs to come from outside the range of the dhcp pools.

Since the purpose of the DHCP is to manage IP leases on a single subnet you only need one server if its pool is large enough. The dhcp pool needs to be large enough to handle all the addresses on the subnet.

I was going to write something about the mask but changed my mind. If eliminating one dhcp server solves the IP problem it should also solve the mask thing too. If not, I'd call a consultant in.

As a general rule, subnet masks are the same on all the PC's on your network. The last one being a 254 instead of 255 should resolve itself with a single dhcp server, unless there are static mapped servers with the 254. Even so it might not be an issue depending on the actual numbers. Subnet mask problems weird. A mask problem indication is if you have sets of PC's and servers that 'see' each other, and other sets that don't.

good luck

[Dancing David]

Sounds liek a mess, unless the manager knows what they are doing. We have enough problems where I work, 25+ building servers, something like 13,000 users and 5,000+ desktops, and one big mama bear server. Each building server has different people building it and maintaining it.

Some buildings are buggy because of the server configuration (printers and machines drop off the list), something happens to the internet rather regularly, and when they have a 'grade' day then wham, the server gets way buggy and slow. The main problem is traffic jams, you can have 2-16 desktops going through one router/dummy switch and maybe one network port. There times where you swear every machine in the network has a student logging on at the same time.

It is rather amazing that it works most of the time.

[Wudang]

Free TCP/IP tutorial and tech ref from IBM. Excellent book
http://www.redbooks.ibm.com/abstracts/gg243376.html?Open