It certainly calls into question the social worth of some data proliferation and yet, at the same time, admits that newspapers are read on-line – where is the NYT data-centre and what is its PUE, its server utilisation, its level of comatose servers, its data storage growth? Whatever we do to improve the energy efficiency of the digital economy we won’t be able to put the lid back on Pandora’s Box; 30% of the global population have access to the internet and the other 70% want it. The digital age has arrived and we all want smart-phones, PDAs, tablets/pads, HD-video at home and 24/7 access to the internet – for good or bad that is the ‘future’ and it is housed in data-centre facilities so James, the NYT reporter, might as well have saved his ‘year long’ research efforts to uncover what we all know and are doing something about – and talking about ‘waste’, what a waste of a years life!
The article (beware, it’s several thousand words) starts with Facebook which is rather perverse as they represent the global ‘greenest’ in terms of PUE and a very high server utilisation (easy when you only do one thing) but also (some may say) a ‘trivial’ load. Rather like Google, they are certainly not representative of the data-centre world that >99% of us work in. Let’s not forget that these social media guys run data centres that ‘waste’ half as much as a normal data centre of 2005 would have done. Or is the NYT suggesting that it is all ‘waste’? They won’t find much sympathy in their readership for that kind of Luddite standpoint unless they are all ‘old’!
Then comes the bit I don’t understand, that “the information industry is sharply at odds with its image of sleek efficiency and environmental friendliness”. Where did that come from? If that were so why would we have The Green Grid pushing its Maturity Model, ASHRAE TC9.9 expanding the thermal envelope to save energy, the EU CoC pushing for both ICT and data-centre efficiency improvements etc. If we didn’t need them why, as an industry, have we created them? Only idiots claim the internet to be ‘green’ and data-centres can be ‘greener’ but never ‘green’. OK, we have the obvious and often quoted carbon-saving applications like video-conferencing and internet shopping instead of going by car to the shops but mostly the growth in data is for personal use and companies having to store more data for longer and do so securely or we bitch about our credit card details being sold etc.
I have presented at many venues over the past two years pushing the audience to understand two scenarios; the low server utilisation rates versus high idle power consumption of some servers (see Spec_Power.org etc) and the effect on data-centre power consumption of governmental desire to offer fast broadband to all citizen – so this is no ‘dirty secret’ as described in the article by an ‘unnamed source’.
One theme that the article bangs on about is the emissions of diesel generators from data-centres. This appears to be an issue in the US where some notable data-centres have failed to get the right permits – but so what? The gensets run for no more than 12h per year even including monthly testing – nothing (literally close to zero) in comparison to the emissions of the typical US automobile population and our cousins still like their huge SUVs! Or take another comparison – domestic air-conditioning is a huge power draw in the USA (but close to zero in most of Europe) and far more than the data-centre population. In a country that depends upon combusting huge quantities of brown-coal in their grid and under-values (and under-prices) energy in the first place the imperceptive tiny effect of data-centre genset emissions the NYT stress is perverse. ‘Nero fiddling’ springs to mind…
Then the article uses the word ‘waste’ and accuses us of not using 90% of the power we draw from the grid due to idling and low utilisation? OK, true a few years back, but the latest generation of 80Plus servers are close to 95% efficient at full load and only use 20% power when idling – the NYT article is rooted in the past – maybe too many ‘unnamed sources’? They also used the consulting firm McKinsey to investigate server utilisation and they came back with 6-12% - I hope the bill was small as ‘everyone’ knows that! Actually it’s not that simple, and, as Oscar Wilde said, “The truth is rarely pure and never simple” (The Importance of Being Earnest, 1895), since it is wrapped up in matters of virtualisation (pushing utilisation to 30-40% and even higher), storage traffic flow and I/O – in other words the server may not be ‘crunching data’ but it is working. I would not disregard the NYT thrust in this area since we have some way to go but the point is that we are doing something about it ‘now’ and not keeping it a ‘dirty secret’. When DCIM matures that will control the server clock-rates in line with compute demand and the ‘problem’ will disappear.
The article also questions the need for energy storage; ‘banks of huge spinning flywheels or thousands of lead-acid batteries’. We are stuck with those since the centralised approach to power quality and continuity is probably more energy efficient and resource efficient than putting the ‘UPS’ function into each server. Certainly the man from EPRI (Dennis Symanski) who was quoted in the article should have pointed out that grid power quality will never (was never and is getting worse) be sufficient to run ICT hardware without PQ enhancement either centrally (UPS) or on-board (a la Google). EPRI know that as they have published the best real data on the subject!
Even ‘security’ comes under criticism by Mr Glanz? For reasons beyond me he seems to think (apparently aided by Mike Manos, whom I know knows better!) that keeping the locations confidential is part of the conspiracy. I don’t want my data stolen, do you? The conspiracy theory seems to be aimed at the government not being able to identify the ‘industry’ and its power consumption? I wonder if any government can track the energy used by TVs, or washing machines, or domestic air-con, or traffic lights? Visibility is an issue but not central to the problem. However, the growth in data centres is simple to predict and Mr Glanz didn’t really need ‘…obtained thousands of pages....’ and ‘conducted hundreds of interviews’ of research to predict that the industry is growing. As described last month in ‘Bitterlin’s Law’ (free, not TM, open to use and abuse!) the growth rate can be estimated as the difference between data-growth in the mobile network (c70% CAGR) compared to Moore’s Law (40% CAGR) less the impact of virtualisation software sales. The result is getting towards 20% CAGR, even in a recession. HD-Video from cameras, smart-phones, tablets and movies are fuelling that growth so maybe the NYT wants us to stop using them – that would be a popularity limiting strategy.
Certainly a paradigm change is required in network photonics to avoid the ‘brick wall’ that is quoted from a former utility executive – again correctly identified but certainly not a secret. There is a great study from a government sponsored research institute in Japan (published on the internet if the NYT will approve you to access it!) that covers the Japanese utility capacity versus their internet traffic and I have been using that for 18 months in presentations in Leeds University and DCD-London.
I could go on and on calling into question the articles premise that all of this is both a scandal and has been kept under wraps but neither is true. So, instead, I encourage you to read the original and make your own mind up. I will only caution you that the article does fall into the usual trap of quoting ‘huge’ numbers to impress the audience. For example a Mr Burton from EMC lets slip that the NYSE ‘produces up to 2,000 gigabytes of data per day that must be stored for years’. A big deal? No, not really, since a 2TB (2,000 gigabytes) hard-drive costs $200 – less than a Wall Street trader spends on lunch! The important point might be the access time versus energy consumed for that storage after the event – and that is mitigated if not solved by using solid-state drives or deep-archiving that is not ‘on-line’ in fractions of a second but rather in a few seconds.
So, the data centre is huge and getting bigger. Data is growing exponentially, but is the only way to slow its growth charge correctly for the power consumed? Energy consumption is huge but declining per unit of compute/storage capacity. The industry knew some time back that we had to improve and we are, both internally (TGG DCMM, ASHRAE et al) and externally (EU-CoC, EN50600, ISO/IEC SC39 WG1).
Maybe the NYT is looking for its own Watergate? Or just trying to a Luddite…