Even a PC connected to the router won't lose its IP address, though yes it will lose contact with the switch and rest of the LAN.
But I think the consensus is to use the (gigabit) switch for switching and the router for routing, so that the OP can avoid that pitfall.

By similar reasoning if relying significantly on wireless one may want extra access point(s) rather that relying only on wi-fi built-in to a router. I know the OP hasn't mentioned it so far but may need to consider things like laptops, mobiles, networked TVs, Blue-ray ...

Edited by prlzx (Mon 05-Sep-11 02:32:48)

As it is the OPs PC then he (she) will normally be aware of the reset having either power cycled or soft rebooted the router and would expect to loose connectivity and have downloads interrupted.

Actually BatBoy is likely correct.

When the router reboots the PC connected to it will likely see the ethernet connection go away, and then come back, similar to if you disconnected the network cable and then reconnected it. Certainly with windows, and depending upon set up other OS's too, will see this and deliberately forget the DHCP lease in case this signifies a change of network.and will send out a new DHCP request once the port comes back up. With most routers the same IP will be issued as the DHCP server running upon it still sees the lease as valid, and reassigns it to the PC. But the PC will lose it's IP briefly.

The fact this will all resolve itself in seconds, automatically, and is therefore no more of a consideration than the temporary drop of connection as the router reboots (the actual reboot will take longer than reissuing the DHCP lease once it has rebooted will).

As has already been pointed out the best practice would be to attach to the switch, so the discussion of this is not terribly relevant anyway.

Assuming that most of your network will be using private IP addresses, probably in the 192.168.-.-. range, why not allocate most of the PCs a permanent IP address? The DHCP server can then deal with phones etc which are more transient.

I ask as this is the way I do it on my 100Mbit network, with a simple switch and reading the above I am wondering if I am slowing things down rather.

Nope, not slowing things down. But not speeding it up much either. Although static IP addresses do tend to be easier to manage long term in a network that doesn't change much.

In reply to a post by philippercival:

Assuming that most of your network will be using private IP addresses, probably in the 192.168.-.-. range, why not allocate most of the PCs a permanent IP address? The DHCP server can then deal with phones etc which are more transient.

I ask as this is the way I do it on my 100Mbit network, with a simple switch and reading the above I am wondering if I am slowing things down rather.

While there is nothing inherently wrong with using static IPs, there is also rarely an advantage that outweighs the extra steps of configuring them.

You would want to confident that:

• you'll never want to connect the device to another network - rules out laptops and any wireless interface (unless part of infrastructure)
• you'll never need to renumber the network (for example to make a site-site VPN link)

In cases where you want something to have the same IP each time (e.g. for port forwarding) DHCP reservations provide for that while still allowing you to see / set all your IPs in one place.

Where you mix static IPs with still having a DHCP server you have to keep notes of the static IPs separately to avoid address conflicts. I've seen a few other workplaces where no-one was tracking them and would have seemingly random connection problems depending just on who was in the office that day.

For example I manage multiple networks and for the office only the routers and servers (DHCP / DNS / file / print) have manual static IPs, the rest are DHCP with reservations for switches, access points and printers.

It's worth remembering that DHCP is about more than just giving each thing an IP address. It also can supply parameters such as the DNS servers to use, the default MTU for the network and less commonly on purely home networks, WINS for computer name lookups, NTP for time syncing.

Say you want to change all devices from the ISP DNS to Google or OpenDNS or to use the router's DNS forwarder, you can change it in one place and it will be picked up on lease renewal.

Nevertheless I can see why people might use them for a very small network (say up to 5 devices).

Edited by prlzx (Fri 16-Sep-11 20:37:19)

Connect to the switch - wire speed.

In reply to a post by prlzx:

While there is nothing inherently wrong with using static IPs, there is also rarely an advantage that outweighs the extra steps of configuring them.

You would want to confident that:

• you'll never want to connect the device to another network - rules out laptops and any wireless interface (unless part of infrastructure)
• you'll never need to renumber the network (for example to make a site-site VPN link)

In cases where you want something to have the same IP each time (e.g. for port forwarding) DHCP reservations provide for that while still allowing you to see / set all your IPs in one place.

Where you mix static IPs with still having a DHCP server you have to keep notes of the static IPs separately to avoid address conflicts. I've seen a few other workplaces where no-one was tracking them and would have seemingly random connection problems depending just on who was in the office that day.

For example I manage multiple networks and for the office only the routers and servers (DHCP / DNS / file / print) have manual static IPs, the rest are DHCP with reservations for switches, access points and printers.

It's worth remembering that DHCP is about more than just giving each thing an IP address. It also can supply parameters such as the DNS servers to use, the default MTU for the network and less commonly on purely home networks, WINS for computer name lookups, NTP for time syncing.

Say you want to change all devices from the ISP DNS to Google or OpenDNS or to use the router's DNS forwarder, you can change it in one place and it will be picked up on lease renewal.

Nevertheless I can see why people might use them for a very small network (say up to 5 devices).

Except when you need to be able to connect to a device, without worrying if the DHCP is working, and has handed it the correct IP address.

Set a static IP, and plug the thing into a switch and you will always know what it's IP is. Use DHCP reservations and plug the device into a switch and you only know it's IP if the DHCP server is working as expected, and every network segment between the device and the DHCP server are working as expected. This makes the DHCP server a single point of failure (and trust me you do not want more than one DHCP server on a network). And this can bring a network down.

Anything upon a network that acts as a server of any description, web servers, DHCP servers, Print servers, file servers, LDAP servers, AD servers, etc etc, should always have a static IP that is independent of any other networking hardware.

The separation of static and dynamic IP's should be a simple matter of setting a DHCP range and assigning the static IP's outside this range.

The changing of network parameters such as DNS should be a sufficiently rare event that it does not need the added benefit of being changed all in one place. The MTU should be handled by ICMP within the network, and setting the MTU for any external connections upon the routers that handle them. This way the default MTU of 1500 is fine, as if any connections to the outside world need a lower MTU this can be set dynamically upon trying to talk to the outside world.

As to client devices (laptops etc) these do not need a static address (if it needs a static address it should be tied to the network, how many file and print servers do you really need to take to another office anyway?) and so they can use DHCP as far as it is available.

If you need to renumber your network for a site to site VPN, and you have a sufficiently large network that this creates a headache you clearly have planned your network poorly. You could use a static mapping of IP's for a NAT device to transparently map your network to for the off site network, and vice versa for the off site network to the local network. No renumbering required. And this can be handled by a cisco PIX506E which although not consumer hardware, is hardly the most expensive router in the world.

I have been involved in the management of a very large network (sufficient number of devices to exhaust the available private IP ranges) that was almost entirely static IP assignments out of necessity. No DHCP server would have been able to manage it, and once you start putting a DHCP server into every network segment, you get into silly money, and lose the advantage of managing IPs in one place.

It is about how well the network is managed. If it is managed well it hardly matters how it is managed, it will work. Similarly if it is managed badly then it hardly matters how it is managed, it will break, a lot.

Also, I use different DNS servers depending upon what I need. Most of my network uses OpenDNS for speed and reliability, some of it uses my ISP's DNS servers as it needs to get proper NXDOMAIN responses and the free OpenDNS service does not do that. I have yet to see a DHCP server that would handle that situation that isn't more hassle to set up than it is to just manage it manually. It's really not a huge overhead.

Thank you both for your very detailed replies, which I am going to read again carefully several times. Mine is just a home network where internal security is not really an issue.

I actually have two routers connecting to two different ISPs on two lines. I add the DNS settings to the routers.

Connected to each router is a linux box, which is acting as a firewall and gets its DNS from the router.

Each linux box has a second network card which is connected to my switch.

(No wireless or DHCP as yet.)

All the computers (up to 6) are also connected to to the switch and get their DNS and internet from one of the two linux boxes.

Also connected to to the switch ( on one of its LAN ports) is an old DLINK wireless router. This can be pointed at either of the linux boxes, but only one at a time and provides DHCP and wireless for the network.

A couple of laptops and up to three phones might connect to this.

It all seems to work fairly well, and when I moved Supplier on one of my connections last week I just had to change the settings in one router.

My laptop normally is normally wired at home and wireless when else where, so does not normally cause a problem.

My printing is all done with a nasty windows share from a machine (mine) that normally runs 24/7.

It all works nicely most of the time, I do know it is not perfect though as

1. when a stranger connected about three weeks ago, I had a master browser issue when he left. Temporarily solved by making sure one of the linux boxes (Samba) always wins the election.

2. Neither of the linux boxes like going to Google news from the links at the top of the Google page. Strangely if you type in a search term on a web search, then when you see the results. click news it does work. News link also works fine from all other machines.

File and drive sharing seems to work well across the network between linux, XP (home and pro) and Win7 (at probably dangerous security levels for an office situation).

Just thought I'd throw in a link to pfSense (http://www.pfsense.org/) if you've not seen it before.

They have just released to 2,x stable with the following features

I wouldn't normally mention it for simple home networks but as you are already using linux for firewalling and have two ISPs it may be of interest. Depending on how many network cards you can fit it may be possible to combine some things into one box..

It's based on FreeBSD rather than Linux but does have a comprehensive web GUI as well as console for setup and monitoring.

I use both Smoothwall Express and pfSense firewalls. The former supports up to 4 network interfaces but the latter is being used with 6 network cards which gives some flexibility on how interfaces are assigned if you have multiple WANs and/or LANs.

pmb00cs makes some good points and ultimately each network will have its own requirements and be subject to resources available to manage it. (And if he has created networks which exhaust 10.x as well as the private parts of 172.y then it's certainly not a trivial matter!).

The short version of what I'd wanted to say is - if people want to be configuring some static IPs manually, it's better to understand if and why they are needed (rather than automatic settings) and keep track of them (rather than just a form of tweaking for its own sake).

For your network how things work will depend on a few factors:

a) whether the two linux boxes use IPs from the same subnet (on the interfaces which connect to the switch), in which case it's a LAN with choice of gateways,
or otherwise you intend to form two LANs each with their own IP range

b) the mode that the routers to the ISPs operate in (NAT, classical routing or bridging) which will determine if you are effectively behind double-NAT - not necessarily a problem for web browsing but other things like FTP and VoIP may complain.

c) the role you want the D-Link wireless router to play - purely as an access point or taking a more active role in managing the network

Edited by prlzx (Tue 20-Sep-11 02:05:53)