Providing Large Scale Cloud Services by Virtualizing Network Services over Software Defined Networks


(T) Internet networks have been designed with two major types of protocols: data plane protocols and control plane protocols. The data plane is in charge of forwarding the packets from end-to-end in the network while the control plane is in charge of knowing the network state and topology in order to assist the data plane to route the packets. The architecture of the data plane is very simple and powerful: Internet applications (HTTP, SMTP, FTP…) run on a reliable (TCP) or unreliable (UDP) protocol that runs on a best effort delivery protocol (IP) that runs on a large number of physical medium (fiber, copper, radio).

The layered architecture of the data plane is the major reason why the Internet has scaled so well to grow where it is today. While the data plane has multiple layers, the control plane (OSPF, BGP, MPLS…) does not. This has led traditional Internet networks to suffer from three major challenges: designing networks is more an art than a science (when it should be the reverse), networks are generally difficult to operate and to manage, and the network infrastructure is hard to evolve.

To overcome, those challenges researchers at Stanford University and UC Berkeley proposed that the control plane should be, instead of being embedded with the data plane in the network fabric, logically centralized or distributed but decoupled physically from the data plane. That proposal led in particular by Martin Casado (the Ph.D. student), Professor Nick McKeown (the professor) and Teemu Koponen (the network engineer) became known as Software Defined Networks (SDNs). Some of the concepts of SDNs have their roots in Darpa Active Networks and programmable switching fabric.

SDNs rely on controllers to provide the control plane functions and communicate to the data plane through the OpenFlow protocol (which must be supported by the networking device through an API). Anyone can define its own control plane protocol using OpenFlow controllers. OpenFlow controllers provide the forwarding tables to the routers and switches via OpenFlow messages. Network applications and management tools can leverage the controllers:


OpenFlow-based Software Defined Networks (picture from Stanford University)

Google, Facebook, Amazon and other large providers of cloud services have been leading the adoption of SDNs running on off-the-shelf hardware. Large datacenters require accommodating on-going changes that SDNs can fulfill.

I attended this week at Stanford a talk to the Silicon Valley SDN group from Amin Vahdattechnical lead for networking at Google. Mr. Vahdat explained how Google is leveraging SDN, to virtualize the key network services such as VPNs, load balancing, access control and Denial-of-Service (DoS) to run Google cloud services around the globe. Google has as well pioneered the use of SDNs for communication between its data centers across continents/WAN. Following is Mr. Vahdat’s talk that he also gave at ONS 2014:

Open Networking Research Center (ONRC) @ Stanford University
Open Networking Foundation (ONF)
Networking Research @ Google

Note: The picture above is the 34th America’s Cup in San Francisco.

Copyright © 2005-2014 by Serge-Paul Carrasco. All rights reserved.
Contact Us: asvinsider at gmail dot com.

Categories: Cloud, Networking