The amount of energy used by data centers is doubling every 5 years, across the world.
Three dimensions of a data center problem: computing power (along with the architecture: mainframe, virtualization...), floor space (cost of location, how to use the available space, layout...) and energy consumption (energy prices, regulation).
Problem: get the handle on ne the numbers (metrics). A few principles: account for systemic losses, link facilities, IT and building, measure the effect of workloads on energy consumption (SPEC to compare vendors)....
Within 2 or 3 years, half of data centers will run out of available energy to expand.
Obsolescence reinforces the problem: floor space with the wrong energy profile, difficulty in distributing energy and cooling...
The hype cycle for data center power and cooling shows a lot of emerging approaches, in the "technology trigger" side (toward the peak of expectations). Most of these are really immature and many will disappear (e.g. dynamic smart cooling). Be careful!
The data center is out of capacity: locate and quantify the problem, assess solutions (consolidation, virtualization, decommisioning...). About 1 year of extra-life can be gained through this process. When it is not possible anymore: build a new data center (expensive), go to hosting company (costly as well) or refurbish (difficult option, when it is live).
Cost elements of a data center: typically people (29%), software (22%), energy (12%). The costs increase and the growth is maximal on the energy (15 to 20% per year).
From a survey done in the June 2009: about half of the respondent know they will have a capacity problem soon or already have. And almost half did have no metrics and are not considering any (almost 20% are using the Green Grids PUE/DCiE metrics).
Start measuring!
Assumption: DC locations will increasingly selected based on low cost and renewable energy sources (35% in 2014). Contrary to the current trend, no one needs a data center in the "tier 1" cities. Other issues for DC location: people and HQ location (mostly for comfort), disaster recovery, network (capacity, latency, cost...).
Containers may be seducing but ask yourself what your benefit is. It looks more like a niche (the example for MS: they have a simple and standardized set of applications with huge scalability). Problems: change; fixing, configuring...
The software part (system and network management) is becoming vibrant. You'll be able to link all the components from the control center, from sensors, with energy consumption at the rack (or even VM) level (e.g. 1E)...
Today, a data center should run at 24 to 26°C. Some are running even hotter. If you are not sure, go by small steps and check everything is right (check for hot spots).
PUE on different levels of data center designs: 1.28 (aggressive), 2.0 (conventional average) and 3.0 (archaic design). 1.4 to 1.6 is a good starting point for a target. But beware of how you measure it (e.g. seasons, day/night variations).
3 design principles
- POD design: incremental methodology with modularized components to get agility.
- Power zones: high / low / medium density.
- Tiering: differentiate for different workloads.
No comments:
Post a Comment