Towards an Open Data Center with an Interoperable Network
Part II – What are we trying to fix?
Over the past several years, progressive data centers have undergone fundamental and profound architectural changes. Nowhere is this more apparent than in the data center network infrastructure. In this post, we’ll take a look at some of the problems with conventional networks, and next time we’ll introduce the fundamentals of an approach to deal with these issues.
Instead of under-utilized devices, multi-tier networks, and complex management environments, the modern data center is characterized by highly utilized servers running multiple VMs, flattened, lower latency networks, and automated, integrated management tools. Software defined network overlays are emerging which will greatly simplify the implementation of features such as dynamic workload provisioning, load balancing, redundant paths for high availability, and network reconfiguration. Cloud networks with multi-tenancy, resource pooling, and other features are becoming increasingly commonplace. Finally, to provide business continuity and backup/recovery of mission critical data, high bandwidth links between virtualized data center resources are extended across multiple data center locations.
Highly virtualized data centers offer greater resource utilization and lower costs. They can also simplify management if network issues such as latency, resilience, and multi-tenant support for public and private cloud environments are addressed. To realize the greatest benefits from virtualization, networks must be optimized to support high volumes of east-west traffic. This can be accomplished by flattening the network to a two-tier design, using Layer 2 domains to facilitate VM migration, and deploying network overlays to enable efficient virtual switches. While existing storage networks will likely continue in their present role for some time, the opportunity to converge networking and storage traffic is enabled by new lossless networking protocols that guarantee data frame delivery. Each of these exercises requires a non-trivial extension of the existing data network. Collectively, they present a daunting array of complex network infrastructure changes, with fundamental and far-reaching implications for the overall data center design.
The networking industry has responded to these changes with a bewildering array of standardized and proprietary solutions, making it difficult to determine the best course of action. IBM believes that the practical, cost-effective evolution of data networks must be based on open industry standards and end-to-end interoperability of multi-vendor solutions (for a few words on the importance of standards, see my last blog entry). That’s why IBM has recently published a series of technical briefs, endorsed by many industry leading companies, that lay out a path towards an open data center with an interoperable network (which we’ll call by its acronym ODIN….after the ruler of Asguard in ancient Norse mythology. Coincidentally, his symbol the valknut looks a bit like a 2 ties network topology).
Next time, we’ll give you an overview of the first series of ODIN documents and discuss why they’re important. Let me know the biggest problems in your network by responding to this post below, or for shorter problems on my Twitter feed.