by Srini Seetharaman
The prevalent migration of application delivery from legacy compute clusters to cloud generates relentless pressure on networking staff for new application needs and push for agility, without a corresponding growth in budget. There is also a stigma of the networking team as being part of the cost-center of the organization. On the other hand, the compute and application IT teams rejoice in being part of the revenue-center of the organization. They talk about “Docker”, “Chef”, “Puppet”, “Salt”, “Ansible”, “Nginx”, “Kubernetes”, “Hadoop”, “Storm” and a whole bunch of DevOps-friendly tools and applications, and how the networking is in their way. Understandably, the job of enterprise network architects and operations engineers has become more complex and challenging.
Software-defined Networking (SDN) proved to be the network IT staff’s key to making their enterprise networks as agile, lean, and programmable as the rest of the IT ecosystem. The first main application of SDN, viz., Network Virtualization, hit the market a couple of years ago and has been instrumental in providing useful abstractions of the network to the compute workload, wherein each workload and its dependencies are encapsulated in a virtual network that is decoupled from the underlying physical hardware. (See Figure 1)
The second application of Network Service Virtualization orchestration further enabled the agility of the network by providing a tighter integration of applications to the network services within a virtual network, thereby making the workload easily portable. As shown in the figure, each application and its network service dependencies work within the virtual network.
Together, the two SDN applications have been vital in bringing all the operational goodness from the computing world into the networking world, and providing a platform to innovate the new ways in which networks operate.
There are several architectural choices in deploying these abstractions in the enterprise data center, and there’s a market full of solutions that are overlay, fabric, or hybrid in nature. From my past experience working closely with a cloud business unit at Deutsche Telekom, it is very clear to me that there is a steep learning curve with the new technologies choices, and there is a significant change in the operational model and thinking. Furthermore, there is a need to quickly adapt to unknown application/business needs; for instance, corporate priorities are pushing for OpenStack IaaS, PaaS, Big Data, Docker support, among many others. This leaves many of us with questions like the following:
- What are the tradeoffs with deploying an overlay, fabric, or hybrid solution for network virtualization?
- Why and how do we deploy virtualized network services? How will it work with existing workloads?
- How do we debug operational issues in overlay/SDN datacenters?
- How is overlay networking used with OpenStack, and with Docker containers?
- How is policy, security, and access control manifested in the new architecture?
This will be the focus of my Network Virtualization, Overlays and Containers tutorial at ONUG Spring 2015 on May 12 -14 at Columbia University in New York City. Look forward to seeing you there.
The Network Virtualization, Overlays, and Containers tutorial will take place from 1:30-4:30pm on May 12th as a part of ONUG Academy at Columbia University. Register now.