Data Center Networking Confusion and Reality

Before leaving for vacation last week, I completed an ESG Market Landscape Report on data center networking. In this process, I worked with a great line up of networking vendors including Arista, Avaya, Brocade, Cisco, Enterasys, Extreme, Force 10, Gnodal, HP, IBM, Juniper, and NEC. Based upon this project, here are a few of my initial thoughts:1. Not so fast on “flat” networks. I’m using the term “flat” networks to talk about non-blocking, any-to-any network connectivity over a single hop. Good idea but do users really need “flat” networks? Yes and no. Yes, they need greater scale and performance with lower latency. Yes, they need data center networks capable of supporting larger L2 domains. Yes, they need redundant links and devices, and yes, they need to overcome the Spanning Tree Protocol limitation around multipathing. All that said, most vendors have 2-tier switch clustering architectures that will be sufficient for many use cases. 2. We need a single standard for data center networks. There is a religious war going on between vendors that support IEEE Shortest Path Bridging (SPB) and those backing IETF Transparent Interconnection of Lots of Links (TRILL). As an industry observer, it is absolutely crazy to me that there are two competing standards here. Judging by my conversations with vendors, there are advantages and problems in each camp so combining these efforts sooner rather than later makes sense. As Rodney King said, “can’t we all get along?”3. Storage convergence. I’m convinced that the combination of IEEE Data Center Bridging (DCB), low latency switches, higher bandwidth, and lower cost Ethernet switches will make storage-over-Ethernet a no brainer. That said, I’m not sold on FCoE. The ultimate winner here may be NAS rather than block storage.4. Programmable flow will play a future role. In a world of cloud computing, scale-out applications, and multi-tenancy, programmable flow will be extremely beneficial. Does this mean OpenFlow? To some extent but it’s too early to say that OpenFlow is the best way to deliver programmable flow services.More soon, I learned a ton working on this project so I have plenty to blog about.