Mastering the Art of the Cisco Kool-Aid Keg Stand

Placing the emerging data center technologies all together

                It seems very appropriate for my final post that I put the emerging data center technologies all together in an end-to-end architecture. In the words of some of the readers, what better way to illustrate how psyched I am about these technologies than with a keg stand - and mind you, no less than a Cisco Kool-Aid keg stand. If you aren't up for the fun, you're going to miss out. On a side note, a huge thank you to Michael for the motivation and Chris to the masterful image!

                While we wait in line for the keg, let's have a quick review of the challenges we're looking to solve. First is the scalability of VLANs and specifically MAC addresses as we densely pack our data centers full of virtualized hosts. With a Nexus 7000 based architecture implementing FabricPath provides the ability to address this by moving from a traditional MAC learning and flooding model to a more robust MAC routing model. FabricPath utilizes Intermediate Systems to Intermediate System (IS-IS) as its control plane providing the fast convergence and scalability needed to handle the work load in the next generation data center. FabricPath is an intra-data center technology at this point in time and helps flatten networks in a safe, sane and scalable way.

                The next challenge to solve is how to provide the ability to move MAC addresses between data centers over a Layer 3 infrastructure. We discussed the considerations with the more commonly deployed L2 extension technologies like EoMPLS and VPLS and are choosing to continue leveraging the Nexus 7000 and the Overlay Transport Virtualization (OTV) feature in a separate Virtual Device Context (VDC) to provide the extension. OTV transparently encapsulates the L2 frames in IP love and goodness before sending them on their way. This provides the Ethernet foundation needed to move IPs between environments that are separated by a Layer 3 network. Depending on what you are doing, say a VMotion between the sites, you'll need to take into consideration the storage component otherwise you are only solving part of the problem. Cisco's partners like EMC with their VPLEX technology and NetApp with their Flex Cache capability have proven they can provide the technology on the storage side to make long distance VMotion a reality.

                Finally the last challenge is egress and ingress routing of the IP addresses that are bound to the applications that run the business. As systems move between the facilities it is important to make sure that we optimize the traffic flows in and out of the data center and minimize traffic on the data center interconnect (DCI). As the challenge is bi-directional, we can divide it into two challenges. The first is egress routing where we can localize our First Hop Redundancy Protocols (FHRPs) such as HSRP and VRRP using VACLs to keep the hellos that form FHRP adjacencies from crossing the DCI.  This provides local egress routing and ensures that only traffic required to cross the DCI does. Routing is handled by the local devices and traffic on the DCI is minimized.

                The second component of the routing is ingress routing. Drawing traffic into the data center where the target system resides only helps performance and minimizes traffic on the DCI. There are a few techniques to facilitate this that include stateful load balancers and NAT, or advertising more specific routes beyond our data center edges where we might be summarizing. There are pros and cons to each but the technique we're focusing on is to change your routing architecture and utilize Locator ID Separator Protocol (LISP). LISP is an IETF draft that has been implemented in multiple pockets of the global Internet and was originally designed to solve Internet scalability concerns. The primary goal of LISP is to separate the IP address from the location through the use of a pull model, similar in concept to DNS that utilizes two namespaces, the Endpoint Identifier(EID) and the Routing Locator (RLOC). These namespaces work in conjunction to allow IP address mobility without creating route table bloat. Using LISP for DCI enables traffic to be routed to its destination specifically without additional overhead on the entire network. Much more about LISP can be found at this URL http://www.lisp4.net

                I say this almost daily, the data center is one of the most exciting places to be today because of the rapid pace of change and the new technologies that dramatically alter the way we architect, design and operate them. I hope you enjoyed my short stint as a Network World blogger as much as I did. It was an honor to be read by such a diverse and technical audience and I appreciate the opportunity. Now if you'll excuse me, it is my turn for the keg.  Keep on chugging!

• Cisco
• DCI
• FlexCache
• Keg
• LISP
• OTV
• VPLEX