The history of computing could easily be defined by the shifting of bottlenecks between processors, storage and the network. Thanks to the rise in server consolidation made possible by virtualization and faster processors, that bottleneck is now generally most acute on the wide area network. In fact, it's all too common for an IT organization to move to save money by consolidating more servers in fewer data centers to experience a rising chorus of complaints from end users about the performance of applications they now need to access across a wide area network.
The use of wide area network (WAN) optimization appliances has mitigated some of those complaints, but there are many instances, especially in the case of custom applications, where prioritizing network access based on a particular network protocol just isn't enough to solve the problem. With that specific issue in mind, Riverbed Technology has been working on a new approach to the way I/O issues are managed across an extended enterprise.
Called Edge Virtual Server Infrastructure, the Riverbed approach is based on a new Granite Core appliance that runs in the primary data center and a new version of Riverbed's Steelhead appliance in the branch office that has been optimized to manage I/O traffic at the block level. According to Miles Kelly, Riverbed senior director of product marketing, that approach allows storage to be decoupled from servers over thousands of miles using a file system technology that Riverbed developed for the Granite appliance. The Steelhead EX appliances essentially accelerate in parallel all the read and writes taking place between the Steelhead appliance and the storage systems in the data center that are connected to the Granite Core appliance. This approach, says Kelly, can reduce the total cost of enterprise IT by as much as 30 to 50 percent by allowing IT organizations to more aggressively consolidate remaining servers in branch offices.
The Granite appliance represents a transformation of Riverbed, says Kelly, in that the company's core mission is to now eliminate performance bottlenecks anywhere they are to be found in the enterprise. That approach has some interesting implications for IT organizations that have been working around performance issues created by shifting I/O bottlenecks for more than three decades.
Those issues have been particularly vexing when it comes to server consolidation. The advent of virtualization has made it possible for IT organizations to theoretically consolidate not just hundreds of servers, but even entire data centers. Up until now at least, the single-biggest limiting factor when it comes to consolidating servers had been I/O performance across the WAN.