George Crump, Senior Analyst

Introducing Scale Out Networking

As stated above scale out networking brings the same flexibility and “pay as you grow” cost advantages to the infrastructure that scale out computing and scale out storage brought to their respective environments. This means that the IT manager can avoid some typical purchasing challenges as their network infrastructure grows. The biggest of these is deciding when to buy a large backbone device and then letting it sit mostly empty. Another is to continue buying individual switches and manage them separately.

Scale out networking is different than simply connecting stand alone switches. While many switch vendors include software that allows their switches to be seen from a single management interface, they can’t be truly managed as if they were one big switch. Users must drill down into each switch, individually, for specific management functions. Rarely, if ever, can a setting change be made that will apply to all switches. Each switch must be accessed, and the change made individually. Finally, the IT manager must know which switch each server is connected to.

With scale out networking the management GUI sees just one virtual switch. The reality that there are multiple switches making up that virtual switch is abstracted from the network manager, although they can drill into individual components as needed. With this type of abstraction the network manager can make global changes to all individual switches that comprise the virtual switch at the same time. While the network manager should be careful not to plug the redundant connections from a server into the same physical switch, they don’t need to be as concerned about which physical switches each server uses. It is true though that the individual elements can become busy, something which can be identified through a ‘top talkers’ type of function and the traffic rebalanced.

There still remains a point in networking where a move to a backbone switch is needed because of the amount of switch interconnects required. The port density on products like the Brocade Virtual Cluster Switch family though, moves that point further off into the future for many companies. When the time does come for a backbone type of investment, users will typically see a higher level of utilization from day one. Since most environments with backbones still use modular switches as the edge part of a backbone-edge strategy, then scale out networking still makes the network management and configuration significantly easier. Essentially, there is only one edge point to manage instead of dozens.

Another objective of scale out networking is to promote the efficient use of available ports. This is accomplished in two ways. Scale out networking should support all the converged protocols, FCoE, NFS/IP and iSCSI/IP, so that more can be done with the same ports. It must also have greater intelligence than what’s being provided by the legacy spanning tree protocol; a technology which effectively deactivates redundant links. Instead, scale out networking products should use TRILL (see our “What is TRILL” article) as a foundation to make sure that all paths are available at all times for maximum bandwidth utilization.

Why Scale Out Networking?

In the future the scale out style of networking we have described above is going to be a higher priority requirement than ever. Network and storage infrastructures, even in the smallest of businesses, are growing faster, driven by a number of factors. First, while server virtualization has made for fewer physical servers to connect, many data center managers are finding that the number of connections per rack is actually higher than before.

Second, storage and network consolidation, as well as bandwidth increases on a per port basis, will alleviate the connection count issues. However, this transition will take time. It’s important that the scale out networking components support IP, and preferably, all the potential converged protocols, like FCoE, NFS/IP and iSCSI/IP. As the transition to a converged infrastructure occurs there may be a swelling of ports until that transition is completed. This could mean that during the transition there are actually more switches to support these dual fabrics. As a result, it’s more critical that management of those individual switches be simplified. Network managers may want to wait until convergence is completed before moving to a backbone device and delay its use even further into the future. Once again, this makes the capabilities of scale out networking and its management even more important.

Third, the ‘drive out cost’ pressures, while intensified over the last few years, will probably never ease in the data center. In addition, concepts like cloud computing and cloud storage have made the idea of buying IT in small, incremental chunks more attractive. Scale out networking allows IT managers to follow that same model with their infrastructure purchases, but without the eventual management ‘brick wall’ that the users and administrators of stand-alone switches will face.

With scale out networking IT managers can scale the network infrastructure incrementally as their needs grow and not make the big backbone investment until it can be easily justified in their environment. It provides the ability to manage multiple switches as if they were one and support both the legacy and newer converged protocols. For growing data centers it’s the logical choice when expanding the network infrastructure.

Brocade is a client of Storage Switzerland