Data Centers and the [Forced] March to an IaaS Utility

Simon Wardley has once again made me think seriously using his “Mapping” methodology as I try and think thru the disruptions in the traditional Enterprise and “Outsourced” Enterprise Data Center(DC) space (whilst managing my own confirmation bias).  We have long used the analogies of electricity and as the “cost of compute trends to the cost of power”, it probably makes some sense to talk about this!Wardley Capability Progression

Most of us know about power generation, power distribution, line loss, co/lo-generation and have even seen “backup generators.” BTW, I believe these to be good analogies as we discuss the evolvement of corporate, CoLo and provider/partner Data Centers.

One critical element in understanding competition in the “production” space is to understand the demand equation.  In power, a huge number of factors govern demand (and also supply) in order to approximate price (Michaillat and Oren, 2007).PowerPriceCompetition

Notice that the price (P) rationalizes a hybrid landscape of generators as well as a calculation that helps to resolve load [uncertainty] over time.  These generators could include long cycle and short cycle (nuclear vs. gas turbine) that help to characterize the responsiveness to demand from low elasticity to “just in time”, and we know that loads will also share the characteristics of permanent, predictable times and unpredictable peaks/valleys.

From a pure math perspective in power-plants there are a couple of key dimensions including capital/fuel/maintenance/loss which are important parts of the calculus, but perhaps most interesting to me is LOSS.  Many people know that power distribution is exceedingly expensive… namely, resistance in the transmission wires/process (~6.1% of net generation), as well as the capital costs associated with the poles, holes, copper, transformers, etc…   I think that there are real analogies to utilities like “compute”.

We, as an IT industry, rarely talk about “loss” associated with distance, and yet these costs are just as real.  Whether we talk about the variable networking costs for ingress/egress traffic (including HA/DR) which might be substantially avoidable by locating closer to the places where information is produced/concerned, or even the costs associated with latency in the form of lost opportunity?.  The traders have been doing this for years, as have outsourcers like CSC.  The real value begins to emerge as colocation begins to create markets like “Big Data Malls” of which a number of us have been discussing online.

Simon Wardley maps emergence of utility markets.

Simon Wardley maps emergence of utility markets.

If we really believe that major industries move to service utilities, as evidenced by the increasing commoditization of computing [checkout  the most recent price wars – apples-to-apples]… Although the move toward analytic services may, as yet, be premature [though I am certain that many of you will tell me of the great analytic services like google analytics and others]

Everything said, data centers are, by their very nature, things that we just want to “take for granted” and this points to a service utility treatment.  These utilities need agreed standards, and in our case PaaS/IaaS via the NIST, to enable reduced friction across the landscape.

James Hamilton on DataCenterWhat OCP has done for infrastructure and OpenDaylight is doing for Networks all in the name of concentrating demand (and therefore focusing supply) needs a parallel in the Data Center. Data Centers, though a useful subproject in OCP, don’t get talked about in an “industrialized” way, but there are a huge number of examples that are very convergent in their thinking [I would be remiss not to mention Amazon’s James Hamilton‘s who has helped me shape some of my thinking

For Data Centers, I contend that the decision calculus include key factors including:

  • stability of capital utilization (and ability to capitalize on multi-tenancy to maximize)
  • peak demands and periodicity of utilization
  • workload friction,
  • network intensities (dimensions of information ingress/egress – volumes, variety, velocities)
  • cost of power, people, capital
  • location of available public / CoLo – IaaS Utilties
  • and, unfortunately, regulation, policy and politics

So lastly,

  1. Is there a need for a common DC pattern? absolutely!
  2. Do you need your own DataCenter? Maybe – depends on the individual economics.
  3. Would you build a corporate datacenter today? depends on multi-tenancy and vectors of information exploitation!

Comments very welcome!

Leave a Reply

Your email address will not be published. Required fields are marked *