Rajat Ghosh is the Chief Executive Officer for AdeptDC, an early-stage startup focusing on optimal cooling allocation for data centers.
Data center managers often cringe at their data centers’ electricity bills and would appreciate a better understanding between demand and supply profiles of their data centers. A new metric—resource allocation index (RAI)—was proposed in line with the philosophy of designing data centers as an integrated value chain rather than a collection of compute and infrastructure systems.
RAI is defined as the ratio of normalized resource supply to normalized resource demand. Normalized resource supply is defined as the normalized electricity used by the data center, i.e., the ratio of electricity used to maximum possible electricity available.
On the other hand, normalized resource demand is defined as the normalized user count in the data center, i.e., the ratio of current number of users to maximum number of users (that can be handled by the data center).
Critical Remarks on RAI
Although the metric has received critical acclaim, there are a few concerns regarding RAI calculation. Two prominent concerns are:
- Since RAI is numerically a ratio of ratios, it is hard to make a business case out of it.
- Most data centers use power usage effectiveness (PUE) as a performance metric. An additional metric means more metering effort.
- The computation of RAI is not straightforward.
Calculating RAI in a Typical Data Center
To overcome these challenges cited by the readers, following modifications are suggested for RAI calculation:
The ratio within the second parenthesis is a veritable constant because it should be defined by the data center designers in terms of IT hardware/software configurations and power source ratings. At this point, it will be useful to map the user count into a related engineering variable such as the data center computing power. User count is often an impracticable variable to assess.
In contrast, the maximum computing power and real-time computational utilization are relatively straightforward parameters to monitor by tapping into the IT management software applications in data centers. The modified RAI is defined as:
The computing utilization or IT utilization is directly related to electricity consumption by IT devices. IT power is equal to the sum of static and dynamic powers. The dynamic power component is directly proportional to the computing utilization.
Therefore, IT power can be modeled as:
Putting back into RAI definition:
This suggests RAI is directly related to PUE. The only dynamic factor for a data center facility is power consumption for computing operations.
RAI Metering Using Infrastructure for PUE Metering
The symbols used in Figure 1 are defined in the following table:
Figure 1 shows the metering architecture for RAI monitoring. Most data centers have this architecture in place for PUE computing.
Clearly, PUE computation does not require resolution of static and dynamic components of the IT power. However, those data are critical for RAI computation.
RAI computation evidently requires measurement infrastructure for IT power. Table 1 (below) informs the various IT power measurement strategies prevalent in data centers. As shown, these infrastructure elements exist in a typical data center. That means RAI computation does not warrant additional capital expenditure.
Table 1: IT power measurement strategies prevalent in data centers
- RAI conveys significant business sense by informing how a data center as a system responding to the incoming computing demand. This is a considerably improved way of metering of a data center as a value chain unlike PUE, which only focuses on the supply-side of a data center.
- RAI can be calculated with the infrastructure existing in a typical data center.
The author would like to acknowledge the contributions of Mr. Mark Monroe, Chief Technology Officer and VP at DLB Associates
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission processfor information on participating. View previously published Industry Perspectives in our Knowledge Library.