Data and the internet are becoming ever more critical to the planet’s residents.
Even the smallest rural villages in Kenya or India are likely using data services on a daily basis; services whose importance can often be placed far ahead of electricity or even sanitation.
The rise of the likes of M-Pesa in what many in the developed world would (arrogantly) consider backwaters is testament to the power of communications and information to transform the lives of all.
No matter whether its an SMS message, financial transaction, email, complex business process or cat video, almost all that we do online will have data being processed at or at least being routed through a data center.
For those who haven’t visited, data centers are typically quite boring places, loud with the sound of thousands of small spinning fans and industrial air conditioners.
Through my early days in the IT field I used to spend inordinate amounts of time in them installing or configuring equipment, I was always glad to leave, usually at 3am.
Data centers are not built as places for humans, in fact, it would be much better if humans needn't venture into them - bring on the robotic server technicians to replace parts and plug cables!
One thing a lot of people don’t realize about data centers is that they use MASSIVE amounts of electricity to run.
While the trend for the likes of home PCs and mobile devices has been to downsize and reduce power, data centers have been the opposite. The servers which sit within are becoming ever more dense and power hungry.
Let’s start with a quick bootcamp. I do understand if this paragraph is TLDR... about how virtually every data centers is structured. Firstly the data center will have a room(s) or hall(s) containing racks into which the equipment like servers actually go.
These rooms and halls will usually be relatively sealed for airflow and have high security access such as fingerprint readers, swipe cards etc.
A standard data center rack is 19” wide (~50cm) and it varies but usually about ~1.2m deep. All equipment such as servers, switches, routers etc. Must be the ~50cm width to mount but the length can vary.
The rack are broken up into RUs or “rack units” on the vertical plane of about 4.5cm per RU. A piece of equipment such as a server is measured in RUs and this is how much vertical space it takes up in a rack.
Usually a server would be 1RU or 2RU but some big pieces of equipment can be 10RU or more. A rack would be usually between 42-48RU high. So if you had 1RU servers, you could fit 48 into a single rack.
A standard server rack. Check the red circles to see how the vertical height system works.
OK now you are a data center expert, let me explain what has been happening in the industry.
Data center space is not cheap, your 1.2m x 0.5m x 1.9m space (yes that is ~1m³) may set you back USD $700 per month without the other mainstays of the data center, internet access and electricity even being considered.
Wow its quite a bit of money compared to your average 1m² of office space!
Alongside this equipment is now more interconnected than ever and there has been a strong push for better utilization of servers.
Previously you had a “server box” running your email, a “server box” running your website etc., what you would find is that each was using 5% of its resources so a big chunk of your RU space was being used by mostly idle equipment drawing not too much power.
Virtualization through the likes of VMware has come along and encouraged the consolidation of these “workloads” as the industry likes to call your website, email etc. Onto less servers with higher utilization.
And now, while you are definitely saving overheads in power use over all the boxes, you have a tighter area using more power than it once did. This has encouraged others to more densely pack their servers and data center space to get the best “bang for buck” on both equipment and space.
To skip a few steps and reach the point, whereas ye-olde 2000s rack may have been using 5KW of power for all the equipment inside, many racks these days may be using 30, 40 even 50+KW of power.
With the birth of the age of AI (artificial intelligence) and huge data processing, with specialized equipment that is coming out, this could get even higher. Imagine 50 fan heaters stacked in 1m² of your house and you can get an appreciation of the power used and heat generated.
In fact, the heat is a big problem, it can take just as much air conditioning power to reduce the heat so it doesn’t cook the equipment or the poor old data center workers.
Now, most data centers I have worked in were relatively small – maybe 100 racks each; I did a quick web search and in 2012 AWS – Amazon Web Services (AWS) had ~5000 racks in one of their hosting regions in the USA.
There’s a recent report from Greenpeace that the AWS region may have 1 Giggawatt (GW) of power at that site very soon and while they won’t be at capacity, they will reach there as they expand – with technology growth a certainty I would claim.
So what does 1GW look like?
According to the Australian Energy Regulator, the average daily power consumption for a 3 person house is 19.2kWh. While there will be spikes in the morning and evening, if you flatten this out that is 0.79 kW on average.
A gigawatt is 1,000,000 kW. So AWS in a single hosting region in the USA has the capacity and will soon be using the power of ~1.2 million average 3 person Australian homes. That is not a small number.
So how do we power data center?
Data centers will almost certainly draw their power from the electricity grid. Grid operators will have specialized lines at high voltage coming into private transformers for the data center.
Most will have batteries and power generators to keep everything operating even if the grid loses power.
Actually, unless you are at Amazon scale where you can probably build your own power grid and power generation, a key consideration of where data centers are built is how easy is it to get enough power from the grid!
And this flows through to the point of this article, location, location, location.
Just some of the power infrastructure for one of Google's data center in Oregon.
One other thing that data center are known for is that they are a hub for connectivity, be it internet, phone, private business network or the like, data cent will have fibre optic cables coming in from all directions; and they are usually the easiest place to get super fast internet connectivity.
In fact, before we had UFB (ultra fast broadband - New Zealand’s highly successful version of the Australian less-than-successful “high-speed” NBN broadband network), the main reason I started using data center was actually more driven by the fact they had fast internet connectivity rather than security or reliability features in the early days.
The nature of communications and the internet means that a data center could literally be on the other side of the planet.
The internet has enabled us, no matter where we are to ubiquitously use any service (well any is not always the case - apart from perhaps those countries with censorship of which the list is unfortunately ever growing).
That being said, we usually want a data center close by as the cost for our ISP of getting the data to somewhere close is much less.
Also the latency, the physical time it takes for lights to pulse down the fibre optics, really has a noticeable effect on speed – a closer data center means lower latency and greater speed. If you want to know more, ask an online gamer, financial trader or someone trying to run a Citrix session to another continent.
So far we have figured out that data center use a heap of power and that is growing and that their location doesn’t really matter thanks to the internet (as long as we have low latency and low bandwidth charges).
Let’s go through some other points framing this idea:
It all just comes together.
In fact on the last point, the aluminium smelter is in one of the coldest parts of the country and has a massive transmission system to feed it already from carbon free hydro generation. It has to be one of the best set of circumstances begging for those wanting to build an eco-friendly and low cost data center.
I know others have spoken about this previously. Lance Wiggs wrote a blog about this 4.5 years ago:
Technology is evolving at massive speed. Take the quoted global power usage of Google in Lance’s article of 260 Megawatts (MW) – that is now a quarter of a single AWS region. Or the single cable system which has turned into 3 within 4.5 years since he wrote the article.
And of course with the ever-growing demand to reduce global warming – its seriously worth consideration.
Lake (and hydroplant) Manapouri - one of New Zealand's largest power stations and beautiful lakes.
One place this may hit some roadblocks is on the legal front. While the techie inside me is always considering the most efficient solution I am cognizant of potential legal issues.
More and more countries are mandating their industries keep data “in-border” after a series of international data scandals and fear.
To me Australia and New Zealand feel in some ways like a contiguous country – visa free travel, incredibly similar cultures (including the Australian vernacular slowly finding its way into New Zealand!), a history of ANZAC, regulations consistent across – the list is huge.
While areas such as the New Zealand and Australian privacy acts differ slightly – so much else can be worked through and hopefully dispensation could be made as part of the countries’ unique cooperation and continuities.
Can we make this happen?
I realize I am a single person making this argument but I am not the only one with this thought train.
In a recent discussion with a major NZ based crypto-miner (crypto-mining takes a HUGE amount of energy to power the computing equipment also), he also was keen on the idea and was far more progressed with trying to make it a reality.
It seems that without an industry and potentially government behind supporting such an initiative it may be difficult to make happen at a large scale.
Actually what kicked things off for me to write this blog was twofold.
Firstly, my AI-based data extraction company, Xtracta is a user of AWS and have been servicing a major new key client from AWS in Australia.
To me, it feels like a backward step providing a service to this client from hot and global warming inducing Australian hosting compared to the clean power and lower temperatures of New Zealand, but unfortunately, there is no AWS in New Zealand.
Can that change?
The other is a large opportunity we are working on at the moment.
It’s ambitious but would see us processing so many documents we are now determining our required resources in the tonnage of equipment and Megawatts of power that it would use.
I am not sure if its commonplace for SaaS companies to price projects like this!
This project is actually a really good example where something like spot server instances from the IaaS (hosting) companies like AWS, Google, Microsoft etc. could come into play since the client doesn’t need real-time processing.
For those who don’t know spot prices allow you to purchase any spare capacity with a hosting company at much reduced rates with the understanding the price for it fluctuates often minute by minute. I am still determining whether it stacks up better for this project.
Perhaps this is a concept the New Zealand tech community can start to raise and get some serious backing behind it.
While I won’t hold my breath for an Amazon, Google or Microsoft to do this (although it would be great to see) – the dta center space is an insanely fast-growing industry.
If we can show good economics coupled with NZ’s clean and green brand I think even alternative hosting or infrastructure providers can make this work. Let’s see.
73% of Australia Electricity is generated by coal-fired plants
Powered by artificial intelligence, Xtracta technology automatically captures data from documents (whether scanned, photographed or digital) and is designed to be embedded into mobile, cloud or desktop applications via an easy to use API.
Utilizing machine learning and Big Data, the technology doesn’t require manual template setup and can scale to a limitless count of document designs - perfect for document types like invoices, receipts, contracts and other high volume documents.
Xtracta processes over 10 million pages a month for customers and is integrated with many well-known software applications including MYOB, Microsoft Dynamics, Sage, Xero and Oracle.