IoT

IoT (Internet of Things) and M2M (Machine to Machine) primarily focus on embedded systems. That is computing and communication facilities integrated on a device, example a household appliance, heating or aircon system etc. This allows for automation, as when all household devices,appliances are connected to the internet there functions could be auto controlled from another device connected to the internet. Example a refrigerator could inventory the status of the groceries (in the refrigerator) and auto order to the local grocery store or send you a list of the required items, also you could verify the status of the groceries in the refrigerator at any time.


IoT is gaining popularity in the Smart Grid, Online Security Services and Retail domains, the embedded systems communicate directly with the computers and thus do not require humans to initiate or control any communications.



Energy Harvesting

Any embedded system requires some amount to power to function and if not directly connected to electricity grid, will run out of battery power in sometime. The embedded device can be designed to extract energy from the surrounding environment, any action which generates power (even minuscule amount)  the embedded system should be able to harness and store. If such a system becomes available than the embedded system could function w/o any requirement of battery or any external power.



Low Power Wireless Communication


ZigBee Alliance


contd..

Software Defined Networking - SDN - Part III

More on  Classification...


Classification can be used to handle L2 forwarding, L2 VLAN's, and conventional IP forwarding, as well as more complex forwarding. All the below is possible:

-Send all VOIP traffic out a specified port on the switch independent of the IP destination address.

-Forward all packets from specific IP source addresses to an MPLS tunnel

-Use the IP source address when choosing a path, this allows IP datagrams from source A to be sent along a different path and from source B along a different path, this is when both source A and source B datagrams are going to the same destination !

-Create classification rule based on the content of the datagram


TCAM Size and the need for multiple patterns




End-To-End Path With MPLS Using Layer2

Software Defined Networking - SDN - Part II

In the last article I mentioned that each network element has an external controller, now using a dedicated controller for each network element would be impractical. 

The network elements are grouped in multiple domains and each domain is managed  by a controller assigned for that domain.

Example if you have 20 domains, than there would be 20 controllers managing these domains and also communicating with each other. Since all controllers are in a sync, a policy pushed to one controller will auto update to all other controllers.

Also one should note that the management application running on the controller does not require microsecond processing, the management applications only run occasionally, example the routing protocols only send updates periodically.


SDN Communication

Controller-to-element communication (OpenFlow)

The OpenFlow protocol is the most popular and widely accepted,  the protocol specifies the controller to element communication method, Item definition,classification and the message format.



Controller-to-Controller communication



Classification Engine in Switches

The data plane in a high-end switch consists of a piece of hardware known as classification engine, A network admin never encounters the classification engine directly, as the classification is hidden in the data plane hardware, eg: when a network admin configures a switch, a software module in the control plane changes the classification rules accordingly.

Now what OpenFlow does is to expose the classification to the network management software, which allows the external SDN controller to change the classification rules directly.



Hardware or Software based classification ?


TCAM (Ternary Content Addressable Memory) And High Speed Classification


If a classification is implemented in software, only one pattern can be checked at a time, i.e if there are N patterns than a software based classification will need to iterate each of the N patterns one-by-one.  A hardware based classifier can perform the comparison of these N patterns in a single step!!


A hardware based classifier implements something called TCAM. A TCAM memory cell
contains logic that can perform a bit-wise comparison and all cells work in parallel to perform a pattern match.

Once a packet is received and is placed in the TCAM hardware, all the pattern matchers receive a copy of the bits in the packet, and they all act at the same time. If a match occurs the pattern matcher  selects the associated action (an integer value), and the classifier returns the integer as the result of the lookup 


Classification Across Multiple Layers

As we can now see that the pattern matching can cross multiple layers of the protocol stack !

Example if you want to send all voice (port 50000) traffic from switch port 25. The pattern will have three components (IP, TCP, Dest Port 50000) in a single pattern.

Now this ability of the classifier to span multiple protocol layers with a single pattern makes SDN more powerful than any conventional packet forward mechanisms.. 

Software Defined Networking - SDN - Part I

SDN is an architecture that is based on the abstraction of the network controls, and which foresees the concept of programmable network... without further ado let's first review as to what is SDN and why we should re-look at our current and future network architecture.



In any commercial network product the internal architecture is divided into two conceptual parts.

• Control plane
• Data plane

Control Plane - The control plane in a network device provides management functions, that allows anyone authorised to perform configuration or any other management task on the device. The control pane functions are implemented in software or on separate embedded processors.

Data Plane - The data plane provides the functionality needed to process packets, and includes network interface hardware, packet ingress/egress facilities and packet forwarding mechanism. Data plane modules are implemented in hardware and are highly optimized.

Control Plane Modules and the H/W Interface

The control plane includes multiple software modules, via each of which the network device could be accessed and configured, CLI,http, snmp are the most common network management interfaces.

Note that the network vendors will provide features on a interface which are not available on other interfaces, this is done to differentiate the vendor products from others and this also results in indirect dependency on the vendor for particular feature set, as configuring other network devices on your network from the same vendor will require those vendor specific features.

Now to support multiple software modules the network vendor creates a "common interface" to perform operations.

The above is what is currently present in your network devices..now what changes 
with SDN...

SDN

In SDN most of the control plane functions are moved to a external controller, an additional control module is added, via this control module the external controller is able to modify and monitor the network.

The management s/w running on the external controller can now directly configure the forwarding table in the data plane. i.e. external controller via the SDN module can now communicate with the "common interface". Now the custom features set available via the vendor's custom interface is no longer in the equation.. i.e. you are no longer dependent on the vendor, now you are in control...

 Now, imagine if you could implement a end-to-end SDN in your network, where the network  devices are just a commodity..think...


In the next article I will go thru the nuts n bolts of SDN with a  how-to implement...