By Professor Jim Al-Khalili - University of Surrey

At the very least, he is the undisputed father of modern optics,­ or so we are told at school where our textbooks abound with his famous experiments with lenses and prisms, his study of the nature of light and its reflection, and the refraction and decomposition of light into the colours of the rainbow.

Yet, the truth is rather greyer; and I feel it important to point out that, certainly in the field of optics, Newton himself stood on the shoulders of a giant who lived 700 years earlier.
For, without doubt, another great physicist, who is worthy of ranking up alongside Newton, is an Iraqi scientist born in AD 965 who went by the name of al-Hassan Ibn al-Haytham.

Artist's impression of al-Hassan Ibn al-Haytham (BBC)
An artist's impression of al-Hassan Ibn al-Haytham - Isaac Newton is, as most will agree, the greatest physicist of all time.

I have been led to believe that Abdus Salam (Pakistani theoretical physicist and Nobel laureate in Physics for his work in Electro-Weak Theory) observed that St Paul's cathedral (by Christopher Wren in 1708)and the Taj Mahal (1632) were made roughly at the same time, and were equal feats of engineering prowess. But whereas the European tradition of theoretical science allowed R&D to flourish, the eastern (Indian? Muslim? Arab?) fall due obsequious patronage. Although this comment referred to an event somewhat later it has stuck in my mind.

Most people in the West will never have even heard of him.
As a physicist myself, I am quite in awe of this man's contribution to my field, but I was fortunate enough to have recently been given the opportunity to dig a little into his life and work through my recent filming of a three-part BBC Four series on medieval Islamic scientists.

Modern methods

Popular accounts of the history of science typically suggest that no major scientific advances took place in between the ancient Greeks and the European Renaissance.

But just because Western Europe languished in the Dark Ages, does not mean there was stagnation elsewhere. Indeed, the period between the 9th and 13th Centuries marked the Golden Age of Arabic science.
Great advances were made in mathematics, astronomy, medicine, physics, chemistry and philosophy. Among the many geniuses of that period Ibn al-Haytham stands taller than all the others.
As commonly defined, this is the approach to investigating phenomena, acquiring new knowledge, or correcting and integrating previous knowledge, based on the gathering of data through observation and measurement, followed by the formulation and testing of hypotheses to explain the data.
This is how we do science today and is why I put my trust in the advances that have been made in science.
But it is often still claimed that the modern scientific method was not established until the early 17th Century by Francis Bacon and Rene Descartes.
There is no doubt in my mind, however, that Ibn al-Haytham arrived there first.
In fact, with his emphasis on experimental data and reproducibility of results, he is often referred to as the "world's first true scientist".

Understanding light

He was the first scientist to give a correct account of how we see objects.

Jim Al-Khalili (BBC)

It is incredible that we are only now uncovering the debt that today's physicists owe to an Arab who lived 1,000 years ago

Prof Jim Al-Khalili

He proved experimentally, for instance, that the so-called emission theory (which stated that light from our eyes shines upon the objects we see), which was believed by great thinkers such as Plato, Euclid and Ptolemy, was wrong and established the modern idea that we see because light enters our eyes.
What he also did that no other scientist had tried before was to use mathematics to describe and prove this process. So he can be regarded as the very first theoretical physicist, too.
He is perhaps best known for his invention of the pinhole camera and should be credited with the discovery of the laws of refraction.
He also carried out the first experiments on the dispersion of light into its constituent colours and studied shadows, rainbows and eclipses; and by observing the way sunlight diffracted through the atmosphere, he was able to work out a rather good estimate for the height of the atmosphere, which he found to be around 100km.

Enforced study

In common with many modern scholars, Ibn-al Haytham badly needed the time and isolation to focus on writing his many treatises, including his great work on optics.
An unwelcome opportunity was granted him, however, when he was imprisoned in Egypt between 1011 and 1021, having failed a task set him by a caliph in Cairo to help solve the problem of regulating the flooding of the Nile.
While still in Basra, Ibn al-Haytham had claimed that the Nile's autumn flood waters could be held by a system of dykes and canals, thereby preserved as reservoirs until the summer¹s droughts.
But on arrival in Cairo, he soon realised that his scheme was utterly impractical from an engineering perspective.
Yet rather than admit his mistake to the dangerous and murderous caliph, Ibn-al Haytham instead decided to feign madness as a way to escape punishment.
This promptly led to him being placed under house arrest, thereby granting him 10 years of seclusion in which to work.

Planetary motion

He was only released after the caliph's death. He returned to Iraq where he composed a further 100 works on a range of subjects in physics and mathematics.
While travelling through the Middle East during my filming, I interviewed an expert in Alexandria who showed me recently discovered work by Ibn al-Haytham on astronomy.
It seems he had developed what is called celestial mechanics, explaining the orbits of the planets, which was to lead to the eventual work of Europeans like Copernicus, Galileo, Kepler and Newton.
It is incredible that we are only now uncovering the debt that today's physicists owe to an Arab who lived 1,000 years ago.

Man doing maths

Anger, relief and disappointment - that's what sums can do
Tears, tantrums and murder. Far from being a cold and rational exercise, maths can provoke the full range of human emotions, explains Professor Ian Stewart.

In these days when wearing your heart on your sleeve is seen as qualifying you to be a true human being, rather than some robotic control freak, scientists and mathematicians are often viewed as being far too rational to be truly human.

Especially mathematicians, who spend all their days doing boring sums in some remote world of the intellect.

Some news, guys: it's not true.

Not just that we don't spend our time doing sums, but also we possess entirely normal human emotions - and express them.

Agreed, mathematicians are seldom seen bursting into tears or shouting in the streets, but that's mostly because mathematicians are seldom seen. Or, more to the point, seldom noticed, because there are hundreds of thousands of mathematically qualified people in British society, working in a huge range of jobs.

Shouting matches

And it's true that the way mathematics is usually presented strips out the emotional element - but the same goes for banking, architecture, whatever.

Anyone who has ever been to a mathematics conference, or sat in a mathematics department common room, notices very quickly that not only are mathematicians emotionally committed to their subject, but the emotions often run high. Shouting matches are not unusual.

The only time mathematics has driven me to tears was when I was 10

There is an important difference, however: when two mathematicians are arguing at the tops of their voices, eventually one of them says: "Oops, sorry, I've just seen why you're right." And the two are once more the best of friends and go off to the pub together.

This reminds me of one of Richard Feynman's anecdotes of a meeting involving some of the world's greatest minds discussing work at the Los Alamos project - he mentions how items on the agenda were raised, circulated once around the panel and then resolved. He was impressed that there was no grand-standing and contradictions.

I also remember sitting watching a particularly vocal disagreement between a number of particularly vocal (male) physicists - what struck me was that women (the few that were their) were excluded form the discussion not only due to the sheer aggression (this in itself did not concern me, perhaps wrongly), but also the allusions were of a masculine & sporting nature and in fact they were treated as if they were not there (I am not referring here to being excessively deferential as men can when confronted by a woman). Certainly, in my industrial career, teams that do not involve a sizeable number of women, is a sure sign that something is wrong, but I digress).

It is wonderful to observe to people reaching an accord through argument. It happens a lot in science, and I suppose in politics and economics, where the rules of the discipline are accepted and respected. I guess in science, the winner stands in stark relief due to falsifiability.

One of the great emotional TV moments for mathematics was John Lynch's wonderful programme about Andrew Wiles's solution to Fermat's Last Theorem, a famous problem that had baffled mathematicians for 350 years.

Relating how his epoch-making solution very nearly collapsed because of a logical error, Wiles is on the verge of tears. The entire programme shows how committed mathematicians are to their research; how solving a problem becomes a kind of personal quest.

Dorothy Parker once said that the movie actress Katharine Hepburn ran "the gamut of emotions from A to B". Mathematicians may not quite manage A to Z, but they get a good way into the alphabet - joy, sadness, a sense of beauty, anger, relief, worry, disappointment.
Maths lecture on the BBC
Hold back the tears

Even a casual glance at the lives of some of the subject's greats should dispel the notion of mathematicians as ultra-rational calculating machines. Leopold Kronecker's dislike of Georg Cantor's new theory of infinite numbers drove Cantor to a nervous breakdown.

Évariste Galois combined dramatic work on the equation of the fifth degree with even more dramatic involvement in French revolutionary politics, culminating in a duel over a woman in which he was killed.

Apparently he wrote up the bulk of his work the night before the dual (with an army officer. Who said mathematicians were smart?) at the tender age of 21.

David Hilbert was incandescent with rage when Kurt Gödel drove a coach and horses through his massive programme to put all of mathematics on sound logical foundations. This was no surprise: Hilbert had devoted years to the project, and had made what seemed to be a lot of progress. Then it all came tumbling down.

The only time mathematics has driven me to tears was when I was 10. There's a lot of evidence that people understand maths much more easily when it is formulated in a social context.

Abstract puzzles involving cards with letters on one side and numbers on the other baffle most of us, but the same question can be instantly obvious when posed in terms of under-age drinking in a pub.

Quite a few psychologists now think that the rational mind cannot exist without an underlying emotional mind

My problem was the exact opposite: I was being asked to solve question of the type: "When Fred is half as old as Emily was when Arthur was born, how many dogs does it take to change a light bulb in three days?"

Write it as algebra, and I could solve it at the drop of a hat. But I was having real problems turning the social story into symbols.

My subject has, however, driven me to the use of distinctly strong language when my beautiful solution that I have been working on for weeks turns out to be riddled with holes.

The emotion of frustration is very familiar to any research mathematician, being the normal state of affairs about 99% of the time. The deep emotion of joy when a solution finally presents itself makes all of the frustration endurable.

Beautiful sunset

There can even be humour in mathematics, and I'm not referring to jokes: the actual maths may be genuinely funny. For example a breakthrough in my current research relies upon dividing both sides of an equation by zero. Ordinarily this is a no-go area, leading to nonsense.

But in my particular problem, you can sensibly divide by zero provided you start not with zero alone, but by two zeros multiplied together. I laughed out loud when I saw how it worked. Mind you, I don't expect you to be equally amused - you have to be emotionally involved in the problem, and frustrated by not being allowed to do what you want, to find it funny.
Calculator
The word 'calculating' has several meanings

Quite a few psychologists now think that the rational mind cannot exist without an underlying emotional mind. You have to be committed to being rational. Only then can you override your fervent desire for certain things to be true, and accept that they're not.

What makes us human is not raw emotion: we share that with many animals. More refined emotion, such as a feeling of awe at a beautiful sunset, is another matter.

But to me, the real essence of humanity is to experience emotions, but not to let them take over completely if that's a bad idea.

Professor Ian Stewart is author of Taming the Infinite: The Story of Mathematics, published by Quercus

I have been doing a lot of iPhone and php work recently (for my sins) and have been using Aptana - one of the things Aptana does is Cloud Computing. I took one of their example projects and deployed it to a cloud - http://compute-aptana-games-demo.aptanacloud.com/.

Although things are quite hokey, still, I think this is a harbinger of Things To Come. Their s a cute little console to see how much resource use is occurring. You can change you cloud size and collaborate - via Subversion nogal - anyone wanting to play email me at aptana DOT play AT korwe DOT com - I am a little busy at the mo, but it is fun.

Incidentally, I am not blown away by Aptana's mobile stuff - though seemingly Nokia is via the WRT toolkit.

A solution to this issue is to use High Performance Computing (HPC) systems, which contain many processors and more memory than desktop computer systems. Many biostatisticians use R to process the data gleaned from microarray analysis and there is even a dedicated group of packages, Bioconductor, for this purpose. However, to exploit HPC systems, R must be able to utilise the multiple processors available on these systems. There are existing modules that enable R to use multiple processors, but these are either difficult to use for the HPC novice or cannot be used to solve certain classes of problems. A method of exploiting HPC systems, using R, but without recourse to mastering parallel programming paradigms is therefore necessary to analyse genomic data to its fullest.

We have designed and built a prototype framework that allows the addition of parallelised functions to R to enable the easy exploitation of HPC systems. The Simple Parallel R INTerface (SPRINT) is a wrapper around such parallelised functions. Their use requires very little modification to existing sequential R scripts and no expertise in parallel computing. As an example we created a function that carries out the computation of a pairwise calculated correlation matrix. This performs well with SPRINT. When executed using SPRINT on an HPC resource of eight processors this computation reduces by more than three times the time R takes to complete it on one processor.

Grid computing technology has long been the darling of cash-strapped academics in desperate need of raw processing power. Now a ground breaking European research effort has created an industrial-strength platform already appearing in commercial applications.

The SIMDAT project has created a portfolio of tools and services that can finally bring the power of grid computing to industrial applications. Grids capture all the resources of connected computers, from storage to computation.

But up to now grids mostly languished in research labs, where they were used to provide massive processing power or to enable large-scale database management. SIMDAT developed essential business functions for grids, like industrial strength service-level agreements, management and security.

For those of you who thought Intel was angling for an HPC play with its upcoming Larrabee processor family, think again. In case you're not a regular reader of this publication, Larrabee is Intel's manycore x86 GPU-like processor scheduled to debut in late 2009 or early 2010. With Larrabee, Intel is gearing up to challenge NVIDIA and AMD for GPU leadership, but doesn't appear interested in exploiting the chip for GPGPU.

Although the company has engaged in some mixed messaging with regard to Larrabee, recent conversations with Richard Dracott, the general manager of the Intel's high performance computing business unit, and Stephen Wheat, Intel's senior director of HPC, has convinced me that Larrabee will be targeted only for standard graphics applications and the more generalized visual computing realm. The latter does overlap into the HPC space, inasmuch as visual computing encompasses applications like video rendering and other types of compute-intensive image processing. But for general-purpose HPC, Intel has other designs in mind.

Angel Lopez posts an updated example of the fractal HPC Server example code

I updated my fractal example to support MPI.NET (Message Passing Interface with .NET) and parametric tasks in Windows HPC Server 2008. The example can be download from my ajcodekatas Google code:

http://code.google.com/p/ajcodekatas/source/browse/#svn/trunk/FractalExample

Nice example of HPC.NET programming that makes pretty pictures — in case you’re easily bored.

I have always wanted to do this but have never had time. Cool. My noddy toy was using mpi on a Tyan 16 node Personal Super Computer. I have a web interface, and a little javascript to set the zoom. It is not a lot of work to get working again, but I am tempted by the Microsoft option.

On Christmas Day, Microsoft published the details on a recent patent application geared toward pay-as-you-go computing.  The abstract details a situation where the supply chain heavily subsidizes the initial cost of computing equipment and in-turn, charges for time and the performance of the machine.  Microsoft notes that the end user could possibly end up paying more for the computer than the original sale price.  They argue, though, that users would benefit by increasing the “useful life” of the machine.

I think that this cloud based computing model is very useful for SMME's and I have a suspicion that it will become increasingly important, one way or another. For the time being, ignore the issue of patenting and think how it can help you, and I am sure that you will see that big installs will become a thing of the past.

The scalable performance level components may include a processor, memory, graphics controller, etc. Software and services may include word processing, email, browsing, database access, etc. To support a pay-per-use business model, each selectable item may have a cost associated with it, allowing a user to pay for the services actually selected and that presumably correspond to the task or tasks being performed,” says the abstract.

The CNN article goes on to describe several other use cases and pricing models.  I’ll let you take a gander for yourself.  Now, this technology doesn’t directly affect high performance computing and we, as an industry, are not mentioned in the article.  However, what's to say that Microsoft does implement such a strategy in their HPC OS platform?  Would there be a market, especially in academia and commercial, where departments pay a fee for expected performance?  How is this different than simply paying per cpu-hour?  Your call folks.

You can read the full article here.

Personally speaking I think HPC and cloud computing will get closer and closer.

Intel Core i7[Duncan McLeod Financial Mail] In 1965, Intel cofounder Gordon Moore famously wrote that the number of transistors that can be placed inexpensively on an integrated circuit doubles roughly every 24 months. But chip engineers could soon come up against the immutable laws of physics.

Intel’s iconic Pentium processor, introduced in 1993, had 3,1m transistors — the tiny switches that process the ones and zeroes of the digital world — and was capable of processing 100m instructions every second. At the time, the technology press was in awe of this incredibly powerful chip, which had 1 500 times the grunt of the first commercial microprocessor, Intel’s 4004 from 1971.

I found the Intel releases for the last 40 years (4004, 8086, 286, 386, 486, Pentium, Pentium Pro, Pentium II, Pentium III, Pentium 4, Pentium M , Core and i7 ) and calculated the doubling time to be 2.13 years. (Some people have too much time.) However I did not normalise it by the die size...

Now, 15 years after the Pentium’s debut, and true to Moore’s Law, Intel has released the Core i7 — and it represents a radical shift in the way the company makes processors.

The i7 is a beast. It has four cores, or processors, on a single integrated circuit. Using “hyper threading” technology, the processor allows multiple computing threads to be run at the same time - so the computer sees not four but eight cores. Operating systems, applications and games are only now starting to take advantage of multi core technology.

This after all *is* the problem.

The new chip has an astonishing 739m transistors and is capable of crunching more than 76bn instructions per second.

Most importantly, it uses a 45 nanometre manufacturing process — the transistors are packed in so densely that light waves can’t enter the space between them. Roughly 400 of Intel’s 45nm transistors would fit on the surface of a single human red blood cell. Or, put another way, 2m of them would fit on the full stop at the end of this sentence.

The new manufacturing process means more powerful chips that run cooler and use less electricity. It also means Moore’s Law will continue to apply well into the next decade.

The Core i7 is also the first Intel microprocessor to use a new material in the “gate dielectric” — a critical component of a transistor. For 40 years, Intel has used silicon dioxide to make these components. But the smaller they have become the more they have leaked current and generated heat — a problem that has threatened to derail Moore’s Law. Under the older, 65nm manufacturing process, Intel has shrunk the silicon dioxide gate dielectric to just four atoms thick!

Instead of silicon dioxide, the Core i7 uses the metal hafnium which has managed to reduce current leakage significantly.

Where next? Intel is already actively talking up 32nm manufacturing and promising chips with 2bn transistors — more than twice what’s in the Core i7. As the technology shrinks, Intel is able to build smaller and less power-hungry chips that can power a new generation of ultraportable devices. It is also able to build more cores into its top-end chips to keep gamers and others who need high-end processing power happy.

Intel says the first 32 nm chips will make their commercial debut in 2009.

None of this comes cheaply, of course. Intel operates seven fabrication plants around the world. Each fab costs several billion dollars. It recently opened a new, 93 000 m² facility in the US — that’s so large, it says, that more than 17 American football fields could fit inside the building. Nearly one-fifth of the space comprises a dust-free clean room — just one speck of dust can ruin a chip.

The tininess of the transistors in modern computer chips is truly remarkable. But Intel and other chipmakers are soon expected to begin running up against the laws of physics. Next year Intel will switch to a 32nm manufacturing process and it says it can see its way clear to building chips using 22nm and even 16 nm processes in the next decade.

Beyond the next 10 years, though, the world will have to move to nanoelectronics, where engineers will literally manipulate molecules to build next-generation circuitry.

"Wow, that's some news this week at SuperComputing 08. Apparently Microsoft Windows HPC Server 2008, with a Chinese hardware OEM (Dawning), made #10 on the Top500 list, edging out #11 by only 600 Gflops. Folks were shocked to see Microsoft getting so serious around HPC; I think we are only beginning to see a glimpse of Microsoft in the HPC field.

From the article

Microsoft on Tuesday hit another high-performance computing milestone by placing its server for the first time in the top 10 on the list of the Top 500 super computers as judged by Top500.org.

Just a year ago, the best Microsoft could do was 116th place based on rankings from Top500.org, which has been benchmarking supercomputers since 1993 with its bi-annual tests it calls "runs."
Download the latest Network World Executive Guide - Mapping a Successful Virtualization Course

Windows HPC Server 2008, a 64-bit system that shipped Nov. 1, came in at No. 10, achieving 180.6 teraflops with 77.5% efficiency at the Shanghai Supercomputer Center and Dawning Information Industry Co.

Despite the high ranking, Microsoft's biggest high-performance computing challenge is likely in front of the vendor – creating easy-to-use developer tools for writing applications for the platform.

In recent months, the concept of 'cloud computing' was all the buzz. European researchers think about another name, the World Wide Grid, which could run on top of the Internet. In an article to appear soon, ICT Results will report about the g-Eclipse project. As the scientists said, 'the g-Eclipse project aims to build an integrated workbench framework to access the power of existing Grid infrastructures. The framework will be built on top of the reliable eco-system of the Eclipse community to enable a sustainable development.' The project started in July 2006 and was successfully completed in June 2008 for a total cost of 2.5 million including a EU contribution of 1.96 million.

From the article

The dream of using the internet to allow people to access as much computer processing and storage power as they need, when they need it, is a step closer thanks to European researchers.

Although a World Wide Grid running on top of the internet is still probably years away from being a reality, the grid, like the web before it, is starting to take shape between academic and scientific institutions.

Where the internet is a communications channel between computers, the grid goes beyond this by not just using the internet for communications but also as a means of sharing computing resources. Every computer and user can access and make use of the combined resources of the grid.

As things stand at the moment there are a series of isolated grids which allow the resources of clusters of computers, at different universities for instance, to be shared. Each of these grids is usually based on its own proprietary middleware which makes interoperability impossible. Middleware is a type of software which connects hardware resources to a grid.

There are different middlewares available, each tailored for different scientific, commercial or industrial usage.

Another barrier to the development of the grid system is its difficulty of use, requiring as it does now knowledge of specialised computer languages and coding skills.

It is against this background that the EU-funded g-Elipse project has been developing an easy-to-use, Windows-like graphical interface which allows access to grid resources with a few mouse clicks.

HP's BladeSystem c-Class server has been named as the world's largest supercomputer. This is the second year in a row that HP has topped the list of the 500 most powerful computers in the world.

HP computers make up 41,8% of the list, with IBM computers making up 37,6%.

The Top500 list is complied twice a year by researchers at the Universities of Tennessee and Mannheim, Germany, and at NERSC Lawrence Berkeley National Laboratory.
HP BladeSystem powers 40.2 percent of the systems on the most recently announced list; this represents more blade installations than all other vendors combined. Versatile, energy-efficient and affordable, HP blade servers provide customers with the maximum density required for high-performance and scale-out computing.
With 201 placements, the number of HP BladeSystem servers on the TOP500 list has increased by 5 percentage points compared to the June 2008 ranking and by 10 percentage points compared to June 2007. The number of high-performance computing (HPC) installations using blade servers on the TOP500 list has increased more than any other single computing architecture. In fact, blade-powered systems are increasingly replacing proprietary systems in the HPC area and legacy mainframe architectures in commercial environments.
"Customers can maximise their high-performance computing investments while increasing energy efficiency with blades, clearly improving their bottom line," says Andrew McNiven, Industry Standard Servers Business Unit manager at HP SA. "The continued dominance of HP BladeSystem customers on the TOP500 list demonstrates the growing market demand for industry-standard architectures that address a broader set of computing challenges at a far lower cost than proprietary systems and mainframes."

Adding cores slows data-intensive applications

Trouble Ahead: More cores per chip will slow some programs [red] unless there’s a big boost in memory bandwidth [yellow].

With no other way to improve the performance of processors further, chip makers have staked their future on putting more and more processor cores on the same chip. Engineers at Sandia National Laboratories, in New Mexico, have simulated future high-performance computers containing the 8-core, 16‑core, and 32-core microprocessors that chip makers say are the future of the industry. The results are distressing. Because of limited memory bandwidth and memory-management schemes that are poorly suited to supercomputers, the performance of these machines would level off or even decline with more cores. The performance is especially bad for informatics applications—data-intensive programs that are increasingly crucial to the labs’ national security function.

High-performance computing has historically focused on solving differential equations describing physical systems, such as Earth’s atmosphere or a hydrogen bomb’s fission trigger. These systems lend themselves to being divided up into grids, so the physical system can, to a degree, be mapped to the physical location of processors or processor cores, thus minimizing delays in moving data.

But an increasing number of important science and engineering problems—not to mention national security problems—are of a different sort. These fall under the general category of informatics and include calculating what happens to a transportation network during a natural disaster and searching for patterns that predict terrorist attacks or nuclear proliferation failures. These operations often require sifting through enormous databases of information.

For informatics, more cores doesn’t mean better performance [see red line in “Trouble Ahead”], according to Sandia’s simulation. “After about 8 cores, there’s no improvement,” says James Peery, director of computation, computers, information, and mathematics at Sandia. “At 16 cores, it looks like 2.” Over the past year, the Sandia team has discussed the results widely with chip makers, supercomputer designers, and users of high-performance computers. Unless computer architects find a solution, Peery and others expect that supercomputer programmers will either turn off the extra cores or use them for something ancillary to the main problem.

At the heart of the trouble is the so-called memory wall—the growing disparity between how fast a CPU can operate on data and how fast it can get the data it needs. Although the number of cores per processor is increasing, the number of connections from the chip to the rest of the computer is not. So keeping all the cores fed with data is a problem. In informatics applications, the problem is worse, explains Richard C. Murphy, a senior member of the technical staff at Sandia, because there is no physical relationship between what a processor may be working on and where the next set of data it needs may reside. Instead of being in the cache of the core next door, the data may be on a DRAM chip in a rack 20 meters away and need to leave the chip, pass through one or more routers and optical fibers, and find its way onto the processor.

In an effort to get things back on track, this year the U.S. Department of Energy formed the Institute for Advanced Architectures and Algorithms. Located at Sandia and at Oak Ridge National Laboratory, in Tennessee, the institute’s work will be to figure out what high-performance computer architectures will be needed five to 10 years from now and help steer the industry in that direction.

“The key to solving this bottleneck is tighter, and maybe smarter, integration of memory and processors,” says Peery. For its part, Sandia is exploring the impact of stacking memory chips atop processors to improve memory bandwidth.

The results, in simulation at least, are promising [see yellow line in “Trouble Ahead].
Photo: Intel

The Future: Intel’s experi­mental chip has 80 cores.

In a conventional data centre, 35% to as much as 50% of the electrical energy consumed is for cooling versus 15% in best-practice "green" data centres.

"Virtually all data centres waste enormous amounts of electricity using inefficient cooling designs and systems," says Paul McGuckin, research vice-president at Gartner. "Even in a small data centre, this wasted electricity amounts to more than 1 million kilowatt hours annually that could be saved with the implementation of some best practices."
The overriding reason for the waste in conventional data centre cooling is the unconstrained mixing of cold supply air with hot exhaust air.

"This mixing increases the load on the cooling system and energy used to provide that cooling, and reduces the efficiency of the cooling system by reducing the delta-T (the difference between the hot return temperatures and the cold supply temperature). A high delta-T is a principle in cooling," says McGuckin.
Gartner has identified 11 best practices which, if implemented, could save millions of kilowatt hours annually.
* Plug holes in the raised floor - Most raised-floor environments exhibit cable holes, conduit holes and other breaches that allow cold air to escape and mix with hot air. This single low-tech retrofit can save as much as 10% of the energy used for data centre cooling.
* Install blanking panels - Any unused position in a rack needs to be covered with a blanking panel to manage airflow in a rack by preventing the hot air leaving one piece of equipment from entering the cold-air intake of other equipment in the same rack. When the panels are used effectively, supply air temperatures are lowered by as much as 22 degrees Fahrenheit, greatly reducing the electricity consumed by fans in the IT equipment, and potentially alleviating hot spots in the data centre.
* Co-ordinate CRAC units - Older computer room air-conditioning units (CRACs) operate independently with respect to cooling and dehumidifying the air. These units should be tied together with newer technologies so that their efforts are coordinated, or remove humidification responsibilities from them altogether and place those responsibilities with a newer piece of technology.
* Improve underfloor airflow - Older data centres typically have constrained space underneath the raised floor that is not only used for the distribution of cold air, but also has served as a place for data cables and power cables. Many old data centres have accumulated such a tangle of these cables that airflow is restricted, so the underfloor should be cleaned out to improve airflow.
* Implement hot aisles and cold aisles - In traditional data centres, racks were set up in what is sometimes referred to as a "classroom style," where all the intakes face in a single direction. This arrangement causes the hot air exhausted from one row to mix with the cold air being drawn into the adjacent row, thereby increasing the cold-air-supply temperature in uneven and sometimes unpredictable ways. Newer rack layout practices instituted in the past ten years demonstrate that organising rows into hot aisles and cold aisles is better at controlling the flow of air in the data centre.
* Install sensors - A small number of individual sensors can be placed in areas where temperature problems are suspected. Simple sensors store temperature data that can be manually collected and transferred into a spreadsheet, where it can be further analysed. Even this minimal investment in instrumentation can provide great insight into the dynamics of possible data centre temperature problems, and can provide a method for analysing the results of improvements made to data centre cooling.
* Implement cold-aisle or hot-aisle Containment - Once a data centre has been organised around hot aisles and cold aisles, dramatically improved separation of cold supply air and hot exhaust air through containment becomes an option. For most users, hot-aisle containment or cold-aisle containment will have the single largest payback of any of these energy efficiency best practices.
* Raise the temperature in the data centre - Many data centres are run colder than an efficient standard. The American Society of Heating, Refrigerating, and Air-Conditioning Engineers (ASHRAE) has increased the top end of allowable supply-side air temperatures from 77 to 80 degrees Fahrenheit. Not all data centres should be run at the top end of this temperature range, but a step-by-step increase, even to the 75 to 76 degrees Fahrenheit range, would have a beneficial effect on data centre electrical use.
* Install variable speed fans and pumps - Traditional CRAC and CRAH units contain fans that run at a single speed. Emerging best practice suggests that variable speed fans be used whenever possible. A reduction of 10% in fan speed yields a reduction in the fan's electrical use of approximately 27%, and a 20% speed reduction yields electrical savings of approximately 49%.
* Exploit "free cooling" -"Free cooling" is the general name given to any technique that cools air without the use of chillers or refrigeration units. The two most common forms of free cooling are air-side economisation and water-side economisation. The amount of free cooling available depends on the local climate, and ranges from approximately 100 hours per year to more than 8 000 hours per year.
* Design new data centres using modular cooling - Traditional raised-floor-perimeter air distribution systems have long been the method used to cool data centres. However, mounting evidence strongly points to the use of modular cooling (in-row or in-rack) as a more-energy-efficient data centre cooling strategy.
"Although most users will not be able to immediately implement all 11 best practices, all users will find at least three or four that can be immediately implemented in their current data centres," says McGuckin. "Savings in electrical costs of 10% to 30% are achievable through these most-available techniques. Users committed to aggressively implementing all 11 best practices can achieve an annual savings of 1-million kilowatt hours in all but the smallest tier of data centres."