Strange Design - /31 on every single interface, no VLANs

[adsyparker]

Has anyone else come across or even seen this type of network before?

I'm designing a network and I've come across a document which is suggesting we use this method of design.

Every (and I mean EVERY) interface is expected to be addressed with a /31 address with the host on the other side also having a /31.

It appears to want to make use of all L3 technologies at the access level.

In my opinion this is very different approach and I have never seen it before.

Any opinions or experience?

Best Regards,

[Jan Hrnko]

Hi Michael,

Is the document available online? I would like to see it.

As far as I know, interfaces with masks of /31 are used with point-to-point connections, such as serial links. I did not came accross a document suggesting such design in access layer, when dealing with host machines. Doesn't the document refer to servers in some way? That would make more sense. But when applied in access layer - you would need a HUGE ammount of routed ports / SVIs and I think that this is very (cost)innefective solution, just to mention one essential drawback.

I am sorry that I could not help you right away, without even seeing the document. Anyway, I am positive that more experienced experts will come and share their ideas and knowledge.

Best regards,

Jan 

[adsyparker]

Hi Jan,

This is an internal confidential document provided by a third party.  Unfortunately for obvious reasons I can't share it.

Best Regards,

[acampbell]

Michael,

This use of 31 bit masks is documented under RFC3021

http://tools.ietf.org/html/rfc3021

Here is another link to Cisco's take on this design

http://www.cisco.com/en/US/docs/ios-xml/ios/ipaddr_ipv4/configuration/15-sy/config-ipv4-addr.html#GUID-9EA50F17-F7AD-4EA2-BEF0-F1A4A8273C4E

Regards,
Alex.
Please rate useful posts.

[snikiforow]

Old Thread, I know.

Yup, Microsoft still doesnt support RFC3021 unfortunately...

Where this could be useful (/31's for all hosts) is in controlling unicast flooding in the event of MAC Addresses aging out.  Specifically this can be an issue with UDP unicast video where the persistent volume of flooded traffic could cause buffer starvation in the switch.

Your flood would effectively be constrained out each individual port on the switch if using /31's (or /30's, although ofc not as efficient w/ respect to addressing)

The more common solution to this issue however would be to setup delivery via Multicast, where the periodic client IGMP traffic would result in keeping the switches MAC table updated regularly.

[Peter Paluch · Accepted solution]

Michael,

I admit I have never heard of this approach before, and definitely, I do not believe it is a sensible approach for common situations.

I agree with Jan. By using /31 netmasks, you are isolating each connected station into its own subnet. I am not sure what is the goal that shall be accomplished by this approach. Each station would require its own VLAN, its own subnet and its own gateway. Routing tables on multilayer switches would bloat and a physical connection or disconnection of any single station would cause a recomputation in the routing protocol, causing a significant churn. DHCP configuration would be almost impossible to perform efficiently, and first hop redundancy protocols would be outright unusable - because even for VRRP that has the least consumption of IP addresses per virtual router, you would need at least 2 unique IP addresses for two routers which you don't have in a /31 subnet.

The RFC 3021 that allows the use of /31 netmasks was created to save IP addresses on point-to-point links. It was not intended to be used on multiaccess network technologies such as Ethernet. It is highly improbable that ordinary operating systems such as Windows or even Linux would allow you to configure a /31 subnet on an Ethernet interface. This design may, apart from other reasons, fail on the simple fact that the end stations won't accept a /31 netmask, voiding the whole effort.

While the idea of routed access layer has been with us for years and is a well-established concept, stations are not put into /31 subnets of their own. Using the /31 netmasks for everything in a switched network is an extremal approach.

My two cents...

Best regards,

Peter

[adsyparker]

Thank you Peter, those are my thoughts exactly.  I am at a loss to understand the thought process behind this.  I was just checking to see if I had missed out on some new 'bleeding edge' design strategy.

I just tested out applying a 255.255.255.254 subnet address to my Windows PC and it refused to let me use the Network or Broadcast address, so you are correct in your assumption.

Best Regards,