March 24, 2013

Déjà vu All Over Again: Can We Please Put SDNs in Context?

What's new under the Sun?  "Nothing" is the historical answer.  "Not much" is pretty close for the current hot topic, Software Defined Network (SDN). 

In some sense, all networks (circuit switched and packet switched) have been defined by software since computers began to control voice switching in the public network more than 50 years ago.  Rather than counting dial pulses and routing calls with purely electromechanical relays, the Electronic Switching System (ESS) stored digits and selected routing connections with digital logic circuits.  

Paths and Policies

In the realm of digital packet data starting with X.25 services, routing and switching have been entirely electronic.  While routers and switches work with different parts of the packet header (at the protocol layer),  both share two prime functions: path finding and policy application.  

Every packet handler operates from a forwarding table, using a list of destinations to which it can send packets.  The list may consist of MAC addresses, IP addresses, or some other element.  The device matches the destination address in an arriving packet to determine where it should exit the device (port)--along with mapping what address it should carry (because the MPLS label, IP address, or Frame Relay DLCI may change).  In a circuit switch the routing table is the cross-connect configuration.

Policy refers to operational conditions that may modify the handling of a packet.  For example, an Access Control List in a router may block a packet.  A frame relay switch will compare the recent throughput to the committed information rate (CIR) for the connection to determine if a packet should be sent or dropped.  Policy in voice circuit switches created the first SDNs on the public network by routing calls based on short dial strings rather than a full telephone number.

Same Principles, Different Implementations

I contend that all networks operate on the same principles.  The differences arise in how a network device builds a forwarding table and how it gets policy information.  For example: 

Telcos prefer deterministic controls that allow operators (or operations software) to make all the decisions regarding routes or paths and every policy.  Examples are MPLS with traffic engineering, permanent virtual circuits (frame relay or Ethernet), and rate throttling for "heavy users" of Internet access.

Internet architects devised open protocols to let the network devices themselves find paths and signal each other regarding policy.  Thus we have routing protocols (BGP, RIP, and OSPF) and path-finding methods that include policy issues (Label Distribution Protocol, RSVP).

Hardware makers used to prefer proprietary methods.  TimeNet "threaded a needle" to set up an X.25 connection while Nortel used its own method to assign a path.  Frame Relay switches all had proprietary ways to find a path to a destination based on a signal from the "caller" (a switched virtual circuit); however, carriers never implemented this feature as a service.

The Packet SDN

Today's packet SDN uses a central processer or server to manage switch and router configurations.  With complete knowledge of the network, the server doesn't rely on routing protocols in every device to find paths.  Rather, the server knows how each device should handle any packet address and pushes a forwarding table and policies to each.  

Network devices can't be completely dumb.  They must recognize and report errors, faults, traffic volume, and perhaps even flow information.  Automatic alternate routing, or failover to a backup path, is also required; this could be managed as a policy implementation or part of the forwarding table set up centrally. Newbridge Networks (in the 1990s) called this concept "center-weighted management."  They applied it to multiplexers (circuit switches), but it's basically the same for packets. SDN or not, each device forwards packets based on addresses, ports, or other fields in the headers --according to rules set up in the forwarding table and their associated policies.  

What's the difference between the SDN and today's Internet?  It's how those tables and policies are set up in each device.  Will it make a noticeable difference?  Perhaps.  

Making it Work

Calculating paths and tables centrally should eliminate route flapping, and network operators will like the ability to control path routing because this helps balance loads across multiple circuits and control latency for the Service Level Agreement on specific connections.  Removing all traffic from a device that needs maintenance or replacement should be faster and easier.  Taking routing protocols off the routers frees up processing power to handle more throughput.  

All in all, not a bad idea--again.


How can you equate circuit and packet switching? Aren't they completely different?

Consider just one bit passing through a switch or router. The bit arrives on a port and must be sent out on another port (in some cases on the same port). The packet switch directs that bit based on information in the header of the arriving packet that holds the bit, comparing it to data in the forwarding table. The circuit switch engine, the time slot interchanger, moves a bit onward according to the cross-connect table that maps each inbound port and time slot to an outbound port and time slot. The bit moves through the switch without knowing how it happened.

Is the concept of SDN not also focusing on making the hardware boxes more interoperable by using a centralised routing and switching instance. So that the proprietary protocols do not force you to stick to one HW supplier across your network. Hence allow to deploy dumb hardware with whatever OAM software layer you as a network operator prefer to use?

Yes, to some extent. Understand that my description of "centralised routing" is only the path-finding function. Data packets or "bearer traffic" does not pass through the central server.
If by "proprietary protocols" you refer to the signaling or control messages, then adoption of an "open" control protocol avoids lock-in. However, not every switch and router will support every transport protocol, for example MPLS-TE or -TP may not be in a node.
Recall my statement that nodes can't be completely "dumb hardware" if they are to monitor and report on data or link problems. How they report should be standardized in the open control protocol.

Search Webtorials

Get E-News and Notices via Email




I accept Webtorials' Terms and Conditions.

Trending Discussions

See more discussions...

Featured Sponsor Microsites



Please note: By downloading this information, you acknowledge that the sponsor(s) of this information may contact you, providing that they give you the option of opting out of further communications from them concerning this information.  Also, by your downloading this information, you agree that the information is for your personal use only and that this information may not be retransmitted to others or reposted on another web site.  Continuing past this point indicates your acceptance of our terms of use as specified at Terms of Use.

Webtorial® is a registered servicemark of Distributed Networking Associates. The Webtorial logo is a servicemark of Distributed Networking Associates. Copyright 1999-2018, Distributed Networking Associates, Inc.