Breakthrough Operational Efficiency Improvement For Data Centers
Poor data center compute resource-utilization is one of the primary sources of waste and inefficiency in computing today. Data Centers often only use only 20 to 40 percent of their available computing cycles, and yet compute resources, including servers and storage represent more than 85% of the electronics cost in a typical data center.
If 60% of compute resources are essentially sitting idle, and compute resources represent 85% of the electronics cost, this means that more than 50% of the electronics cost in the data center is sitting idle. This represents waste and cost on a massive scale. In a web-scale data center this can translate to hundreds of millions of dollars in wasted CAPEX.
Even a small increase in compute-resource utilization translates into a huge savings in total data center CAPEX. An improvement of just under 10% is equivalent to saving the entire cost of all of the networking equipment in a typical data center.
CALIENT’s LightConnect™ Fabric is a new approach to improving server and storage utilization rates in data centers. It achieves this by allowing Pod resources to be flexibly shared and reassigned at the optical layer in response to the needs of workloads.
This is achieved through a new virtual Pod or V-Pod architecture. In a typical V-Pod, some resources are static while others are dynamic – connected through OCS-enabled LightConnect™ Fabric paths to the spine switches rather than directly or through manual patch panels.
The static resources are permanently assigned to each physical pod while the dynamic resources can either be used within the physical Pod or allocated to other Pods. So, the dynamic resources can be considered to be resource pool available to the whole data center.
As workload-driven compute demand grows within a Pod it can be reconfigured to borrow resources from other Pods. Depending on the granularity of the LightConnect Fabric, it can borrow rows of racks or individual racks.
The power of this approach is that it is no longer necessary to over-build every Pod with resources to handle worst case anticipated demands. Instead we can overbuild by an average and lower amount across the entire data center and reallocate this on demand between Pods.
As a result Pods can be operated at much greater capacity and operational efficiency and this translates directly to lower compute resource CAPEX.
Customizing The Mix of Compute Resources to Match Workloads
CPUs, GPUs, FPGAs, storage, memory, and other compute resources can each be optimized based on the V-POD concept. For example, it is possible to create customized compute capability by combining server CPU resources from one physical POD with GPUs, storage, and accelerators from other physical PODs to create a virtual POD that meets a specific workload requirement.
The LightConnect™ Fabric Manager
The LightConnect™ Fabric physical data-plane layer is built with a network of CALIENT’s S-Series Optical Circuit Switches, managed by the LightConnect™ Fabric Manager.