Data centre cost #1 - Chargeback metrics

Many organisations are now under pressure to develop methods of charging the consumer of ICT services for the data centre costs. These range from colo providers changing their billing structures to include power multipliers through IT departments who have been given their data centre power bills to large enterprises trying to holistically understand ICT cost allocation.

To explore some of the common pitfalls in chargeback metrics and why multiplying power by PUE doesn’t work we can use an analogy to explore fixed and variable costs.

Consider that I work part time for a company (much like a server) and they want to understand if I am worth the cost. To do this accounting need to understand what it costs the company to have me working for them so that they can bill my line manager’s cost centre. This should include not just my salary but all the other overhead costs of employment.

Fixed costs
Now, my manager doesn’t like any of his employees out of his sight whilst we are working (a server hugger) and therefore I am allocated a desk in the office. This desk uses space in the office, which includes some of the lease cost of the building, furniture, cleaning, security, lighting etc. and must therefore carry some cost related to the desk and space. The key point is that these costs are fixed, because my manager won’t let anyone else sit at the desk allocated to me (just like the space and power provisioned to a server installed in the data centre) this cost needs to be applied not just on the 3 days a week I am in the office but also on the rest of the days when I am not in the office but nobody else can use the desk.

Note: I will try to pick this topic up again in a post on supply versus demand side incentives.

Variable costs
The rest of my costs are somewhat more variable and depend on how much time I am at work for and how hard I work.

Direct variable costs
There are some direct costs that map onto the time I am actually in the office, mostly my salary which is an hourly rate. This can be compared to the power consumed at the server plug which may be directly metered and reported in the same way that my hours are recorded on a timesheet.

Indirect variable costs
Once I am in the office and working though there are a number of additional variable overhead costs which the company incurs. Part of my work is payment collections and therefore I make telephone calls, many of them long distance or international. When I am at my desk my PC is turned on and using power, I also breathe and load the air conditioning with waste heat and moisture which needs to be removed. These sum up to the indirect variable costs of me being in the office and can be compared to the additional losses in the data centre power and cooling infrastructure that result from a server drawing power and giving off heat.

Chargeback
Now that accounting understand how my actual cost to the business is built up they can charge my line manager’s cost centre in a realistic and defensible way;

1. Charge the fixed cost of my desk on all days, whether I work or not.
2. Charge the variable direct cost of my salary for the hours I work.
3. Charge my share of the variable indirect costs of phone bills, power and building air conditioning for the hours I work.

Data Centre Chargeback
Now that we understand the difference between the fixed and variable costs we can apply this to the data centre and start to understand what the real power cost of an IT device is.

1. The “Server Power” drawn at the plug is the Direct Variable cost
2. The “Variable Overhead” losses in the data centre power and cooling infrastructure based on the power drawn and heat given off are the Indirect Variable cost
3. This server has 600W of power and cooling capacity assigned to it based on its PSU rating, meaning that its “Fixed Overhead” is the Fixed cost which occurs whether the server actually draws power or not

Multiply by PUE
Now that we have the overall server power at the utility we can compare this with PUE based estimates of chargeback and see whether they are useful.

The basic approach is to determine the PUE of the data centre by comparing the total utility power draw with the IT power draw and calculating a PUE. Unfortunately the PUE of a data centre changes based on many factors meaning that an approximation is generally used. This might be the “Design PUE” of the data centre or a measured PUE. The “Design PUE” being the PUE that the data centre design might achieve given optimal circumstances (such as being completely full with equipment at full electrical load on a cold day) and the measured PUE being determined by the metering system at the time of measurement.

Neither of these methods is capable of representing the fixed power costs of the data centre infrastructure and therefore start at zero when this example server is actually responsible for almost 400Watts of fixed power cost before we add the server power and variable overhead.

Summary
To sum up, we could just multiply my salary by some averaged accounting ‘burden’ factor (like the data centre PUE) to try and get an estimate of what I cost but this is not a very useful approximation of my cost and fails to provide any incentive for my manager to use shared resources more efficiently. In the same way in the data centre the design PUE is far too optimistic and never represents the true cost of the server whilst the Measured PUE is much like a stopped clock and only right for one value of server power draw and neither provides any incentive to manage the allocation of capacity.

An effective chargeback metric needs to capture both the fixed and variable costs of our IT devices to properly reflect the financial or energy impacts of different decisions. In reflecting these costs we also need to be able to account for both the capital and operational costs of the data centre or we will not be able to usefully represent the cost or value of the data centre to the business.