Data Center

Overheard - Hot aisle / cold aisle containment

In any push air-conditioning arrangement, two components need the most amount of power: the compressor and the fan. In a hot aisle - cold aisle arrangement, the need for cold air is reduced drastically because it’s needed only for the servers and not the entire room. Smrutiranjan Das, Hot aisle - cold aisle design best practices for data centers

Today’s WhatIs.com Word of the Day is hot aisle - cold aisle contaiment.

PUE - Power usage effectiveness

A typical data center facility spends almost half of its energy consumption on the systems powering and cooling the computers inside — and not on the computers themselves. Google, Efficient Data Center Summit

Last October, Google disclosed details about its data center energy usage, saying it was averaging a Power Usage Effectiveness (PUE) rating of 1.21 across its six company-built data centers. To put that into perspective, a PUE of 1 means every watt goes to computing machines. A PUE of 1.5 means that half the power goes to non-computing functions like cooling or lights.

Six months later, Google says it’s got that number down to 1.19.  So how is the search giant doing it? For a long time, nobody knew.  Google’s infrastructure was top secret. But then in early April, Google held a summit and gave everyone a peek behind closed doors. Surprisingly, what they seem to have done was follow the KISS principle. Relatively speaking, they are keeping things simple.  They have their own proprietary servers and the data centers themselves are just about as lean as you could imagine. Google’s made several tours available on YouTube — and this one (below) of a data center built out of shipping containers is just amazing.  There are over 45,000 servers housed in 45 containers.  Talk about utility computing — this container tour sure looks like a utility plant to me!

At the summit, Google reps shared best practices, saying:

The best practices we’ve presented here are the main reason we’ve been able to achieve our PUE results and can be implemented in most data centers today.

• Measure PUE (Circuit transformers)
• Optimize power distribution (High efficiency transformer and UPS)
• Manage airflow (Close-coupled cooling — Eliminate hot/cold mixing)
• Adjust thermostat  (Raise cold aisle temp)
• Use free cooling  (Chiller bypass, water-side economizer)
• Use free cooling  (Chiller bypass, water-side economizer)

Overheard - Virtualization sprawl is like ‘The Trouble with Tribbles’

“When we started out with virtualization, we thought, ‘VMs are easy and cheap, so let’s go. Let’s create as many as we need.’ We ended up with several hundred more machines than we actually needed.” Chuck Brust, as quoted in Virtualization users beware: Sprawl is real

Today’s Word of the Day is virtualization sprawl.   It wasn’t such an important consideration when VMs were just used for testing, but now that we’re finding VMs in production, vendors are scrambling to come up with tools that allow administrators to monitor and manage VMs more efficiently so that virtualization remains cost-effective. The holy grail? A tool that allows the admin to see and manage both physical machines and VMs from one dashboard.

Every time I read about virtualization sprawl, I’m reminded of the Start Trek espisode The Trouble with Tribbles. Click - new server.  Click- new server.  Click - new server.  Will the exponential propogation of VMs will eat any profits we might have gained from moving to virtualization?

Overheard: There’s money to be made building data centers

Data center construction is probably the only healthy segment of the building industry. Michael Kanellos, Data Center Power Consumption: By the Numbers

Overheard: Your PUE number is like golf

“Your PUE number is like golf — the closer to 1, the better. At least that has always been the common wisdom. The goal, says experts, is to reduce your PUE. But sometimes an IT energy efficiency project can play games with that number.” Mark Fontecchio, How a virtualization and server consolidation project could hurt your PUE

Overheard: Are data Center efficiency metrics purposely obtuse?

I have come to the realization that this industry does a wonderful job in telling its members WHAT to do, but lacks to follow-up with the HOW. Michael Manos, Struggling with CADE, McKinsey / Uptime Metric

Lots of buzz about the Uptime Institute’s symposium on IT Energy Efficiency and the study McKinsey released called Revolutionizing Data Center Efficiency . It’s interesting that Michael Manos estimates that only 10% of data center managers measure the efficiency of their facilities. Could it be because it’s almost impossible for the average Joe to get the data they need to plug in the formulas? Or could it be that they’re waiting for a clear winner in the proposed metrics? Try this one out for size: CADE.

CADE is the new metric-of-the-week.

CADE (Corporate Average Data center Efficiency) = (Facility Efficiency) x (Asset Efficiency)

Facility Efficiency is defined as (Facility Energy Efficiency) * (Facility Utilization)
Asset Efficiency is defined as (IT Energy Efficiency) * (IT utilization)

Overheard: Is there a market for the heat created by a data center?

The designers of the new mall worked out a plan to connect their heat pump system to the data center of a near by Telco facility. During the winter, the data center heats the mall and during the summer, the mall uses the heat generated by the server room to preheat a local hotel’s water system. kstroh, Selling Heat

Mercursenteret is a shopping mall established in an old telegraph building in Trondheim, Norway. The shopping mall is cooled and heated by a heat pump system. In the winter during the heating period, the heat pump uses the cooling distribution system of the telecommunication centre in the next-door building (Teknisk Telehus) as the heat source. In the summertime, the heat pump operates mainly for cooling. Excess heat from the heat pump system is sold to pre-heat sanitary water in a hotel in the neighbourhood. The seasonal performance factor is 3.5 for heating and 2.5 for cooling.

Tour: Server room at the Sudbury Neutrino Observatory

Tour: Huawei Data Center

Tour: WickedInk takes you on a tour of his data center