In the constant effort to increase cloud and data center flexibility and processing efficiency, the concept of application-defined networking is the logical step beyond software-defined networking for handling applications efficiently over the network. But the implementation of ADN is a work in progress, as the technologies needed to support the concept emerge and vendors roll out their versions of ADN-enabled products.
The concept of
Software-defined networking promises to utilize network resources more efficiently than traditional switching and routing protocols by configuring individual packet flows. And while optimal packet flows are clearly important, an increasingly important purpose of networking is to support application processing.
But processing applications efficiently requires more than optimizing individual flows, because complex applications consist of components that are executing on different servers. It requires multiple network connections, each with different requirements for bandwidth and delay characteristics. Clearly, configuring servers and the network manually to accommodate a set of applications is no longer sufficient.
Applications are evolving constantly to meet rapidly changing business requirements, and a large number of applications are supported in both public and private clouds. Having to manually configure a cloud to support a small set of applications would negate the benefits that cloud processing offers.
With the information gathered from applications, ADN is engineered to achieve the following benefits:
- Network management software and tools can balance traffic loads across available links to provide the necessary bandwidth, latency and delay characteristics.
- An interface to server management software can help balance compute loads across servers.
- More efficient use of network and server hardware will remove the need to over-provision the network to ensure adequate resources.
Different vendor approaches to ADN
Application-defined networking remains a relatively new concept, so a widely accepted definition of the service it provides or how those services can be implemented hasn't yet been crafted. As a result, vendors are adopting differing approaches. Some are enhancing existing products, while others are developing new ones.
In one approach, application delivery controller vendors including Brocade Networks, Citrix Systems and F5 Networks are extending the load-balancing capabilities of their products. Application delivery controllers initially served as load balancers, but they have visibility into all the applications using their services. Interfaces to network management software and tools in recent products enable them to communicate application requirements for network resources.
More application-defined networking resources
Application-aware networking: The fundamental guide
In SDN, applications define the network
Next-gen application delivery controllers enable cloud app delivery
In another approach, Cisco Systems recently announced its Application Centric Infrastructure initiative that would enable applications to run in any environment, cloud or traditional data center, virtually or on bare metal. The goal is that users will be able to specify application requirements via open interfaces, and the infrastructure will automatically supply network, compute and storage needs. Cisco's initiative is expected to include common policy management across network, security and applications, plus updated switch hardware. The initial components of Cisco's initiative will be released in the second half of 2013.
Other vendors are developing software-only ADN solutions. Lyattis Inc., a recent startup, has released two products -- CloudWeaver Discovery and CloudWeaver FlowMapper. Both products are available as a service on Amazon Web Services (AWS). CloudWeaver Discovery locates and reports on allocated AWS resources and their geographical locations, while CloudWeaver FlowMapper displays currently active application data flows, reports on available bandwidth and latency and highlights developing bottlenecks. Currently, both products display information but do not re-provision the network or servers, so operator intervention is required. But in the future, Lyatiss envisions an ADN platform that automatically allocates cloud resources.
- Boundary offers a Software-as-a-Service solution that identifies application components and the flows between them. It constantly monitors those flows and reports on bandwidth and other critical parameters. It supports both public and private clouds and traditional data centers.
- Plexxi's product consists of Ethernet switches and a controller. Switches are connected to each other by dual optical rings, which flatten the network by eliminating the need for a hierarchy of switches. The server-based controller directs the switches to allocate network resources to meet application requirements.
Combined, the Boundary and Plexxi products automate the feedback process. Boundary detects application components, continually monitors the flows among them, then interfaces with the Plexxi controller to direct switches to provide adequate network resources. This solution requires the use of Plexxi switches and controllers, so it can be used only in networks based on those products.
The concept: The approach to combine monitoring software with an SDN controller can be extended to other vendors' equipment. The Plexxi controller-to-switch protocol is proprietary; it does not use OpenFlow. But monitoring products like those from Lyatiss and Boundary or similar products could interface with any SDN controller supporting OpenFlow. A variety of OpenFlow-compliant switches are now available, so the solution could be applied widely.
Application-defined networking's model is new enough that only a few products are currently available, and those are in their early phases. But more products are sure to come. As ADN is implemented and enterprises and vendors gain experience, products will be refined and standard interfaces will be developed. Application-defined networking may then become a widely adopted concept used in every cloud and large data center.
This was first published in August 2013