With the business and application needs described above driving data center architectures, we see a new wave of data center networking requirements emerging. The following key requirements were developed by the SDxCentral research team based on detailed conversations with our readers at enterprises and cloud service providers, as well as our own research into key data center trends:
Simplification, Standardization and Modularity
To meet the drive towards large scale support for applications and data, the next gen data center network (NGDCN) will need a new level of simplification and standardization. Historically, data center networks had multiple different classes of proprietary networking gear, each with highly differentiated capabilities: e.g. L2 devices with limited L3, L3 devices and specialized L4-7 appliances.
Data center operators are moving towards reducing the number of complex routing/switching elements and leveraging standardized compute servers to perform networking optimization (aligned with SDN and NFV trends). Likewise, many NGDCNs tend to focus on simplifying the network fabric with use of fewer commodity chipset variants (but ensuring they have enough switching capacity to meet traffic needs). This is often coupled with a migration of L4-7 functions into standardized server platforms.
As we’ll discuss later, the NGDCNs found at some of the world’s most advanced data centers (LinkedIn, FaceBook, Google, Microsoft) are driving towards modular pod approaches, each of which contain a standardized network architecture (usually some type of Clos fabric) utilizing cheaper commodity switches based off switching merchant-silicon from providers like Broadcom, Marvell and a few up-and-coming networking chip makers (Centec, Barefoot and Cavium Xpliant). In many of these large cloud providers, their switches are custom-built for them by major original design manufacturers (ODM), and recently, these standardized designs have become part of the Open Compute Project (OCP) hardware (HW) project standards.