Take the 2-minute tour ×
Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

I had an idea for designing a network setup in order to acquire high availability connections to my (small) network, but I seem to be running into a few problems trying to find out the particulars.

The Quest

I want to be able to have, in a location where dedicated circuits are at best infeasible but commodity connections of many different types and sizes are available, the ability to have a network that is considered to be “highly available” in the face of connection failures to the upstream network(s). I currently manage a few networks where there are two sets of IPv4 address space, and it is a bit of a PITA to continue doing it that way.

I think that I have found a possible way to work around that, and get something closer to a “normal” network (in terms of appearances, anyway). But I am having trouble with the particulars.

Requirements

For a first run, I would need:

  • Two independent connections to commodity ISP services. Already done.
  • A router capable of bonding two tunneled links together. Already done.
  • A portable routed /28 network block. I cannot find any place that will sell a network block of this size to me, without also constraining it to their network. If they have a multiply-redundant network that I can host on, then I am happy. Otherwise, I can’t use it.
  • A server running on the Internet, preferably in a data center (and even a VPS would be acceptable for a test run), which can “host” the network. I was hoping to do an initial test with Linode, but they said that they require 100% address utilization and only assign addresses individually; they won’t sell you a routed block, nor will they let you bring your own.

The idea would then be to have the system in a data center have the ability to tunnel to any endpoint that authenticates itself to the server. All such tunnels would be “bonded” together so that they can provide both aggregate bandwidth and durability. The router on my own network would be connected to each of the ISP connections, and initiate one tunnel connection to the data center system for each ISP connection. (Also note that the router will handle IPv6 routing directly to the Internet, without having to use this somewhat convoluted setup.)

Can it be done?

So far, it seems the two questions here are blocking my path to success here:

  • Where can I buy a small (say, a /28) IPv4 network block that is portable?
  • Where can I rent a VPS that will let me bring my own IP block?

Thanks for your help!

share|improve this question

1 Answer 1

To be reachable you need your prefix to be visible in the DFZ (Default Free Zone). That basically means that all companies and service providers must accept your prefix and then announce it to their neighbours, customers and upstreams. Almost nobody accepts prefixes longer than a /24, so getting a /28 routed globally is virtually impossible.

What you seem to want is LISP. With LISP you can get addresses (IPv4 and/or IPv6) from your LISP-ISP and route them to wherever you want them. You can use multiple connection from different ISPs. Those connections and their IP addresses are then only used to be able to reach the LISP routers of the LISP-ISP, and your internal network can use the LISP addresses without being dependent on the connections. As long as one connection is working your LISP addresses will be reachable. Redundancy and flexibility without renumbering :-)

share|improve this answer
1  
+1. This basically is "I want to run my own car plant for 5 cars a year, but it makes no sense, help". It as long decided that /24 ist he smallest routable part. Get used to it - small independent routing is OFF. Nice try, makes no sense. –  TomTom Mar 5 '12 at 14:40
    
So it isn't possible to have a BGP advert for less than a /24, that makes total sense. I suppose the answer would be that I would have to provide some sort of service for other small business entities to be able to use, as well. –  Michael Trausch Mar 5 '12 at 15:39
    
The rest of your setup doesn't make sense either. The configuration is complex, you will have problems with the load balancing, failure detection, etc. You make yourself dependent on the data centre where your server is hosted, and servers are a very important single point of failure. Don't try this kind of stuff if you don't have a lot of experience with routing. In trying to make it better you'll end up making it worse... –  Sander Steffann Mar 6 '12 at 17:02
    
The kernel handles the load balancing (though in the normal case, the benefit is little compared to the benefit of transparent fail-over). While I am tied to the DC as a SPOF, DCs tend to be more reliable than the networks I am looking to improve by leaps and bounds. A VPS system can also go down without much impact; it can be replaced nearly instantly if necessary. Same goes for a server inside a data center if there is a warm standby (which there would be). Additionally, this would be one case where spinning drives are utterly useless. :) As to configuration, it'd be DRY'd with Puppet. –  Michael Trausch Mar 7 '12 at 19:09

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.