Persistent Visions

The Microsoft Cloud Computing Strategy
Prakash Sundaresan and Alex Chan

I attended this talk because it was sold as a talk about cloud computing and the challenges, future direction of cloud computing. While I did gain some information about cloud computing, the talk was unfortunately not very informative on the whole. At various points in the talk I wondered if I was at the right place since most of the attendees looked like they were there for the networking part of the “talk and networking session”.

Nevertheless, some points to bring away from the talk:

• Cloud computing is extremely popular in China, where small and medium businesses (SME) cannot afford the capital to set up their own servers and IT departments. This is similar to the popularity of cell phone technology, which bypasses the need to put down land lines. Makes sense to me.
• In contrast to the traditional local server system with expensive/reliable server and related hardware, Microsoft is trying to piece together their system from cheap, standardized units. They then use redundancy and software to self-repair the system when one unit fails. When they reduce the costs, they can sell their services at competitive prices.
• As of now, you can really write your piece of code, upload it to their cloud computing server in Singapore, and let it run — for a certain charge of course.
• Many data providers (both free and commercial) are on cloud computing because it allows them to easily distribute and charge for their data.
• Examples of cloud computing applications:
  • Eye on the Earth by the European Environmental Agency
  • Be a Martian by NASA

One thing I kind of disagree with is how cloud computing is presented as the natural development to the server/client system. I think it that the development depends on the creation of widespread, publicly available network infrastructure. That is, while cloud computing frees the clients from having to spend capital on infrastructure — but someone is paying for the infrastructure whether the someone is cable companies, internet providers, or taxpayers. Paying Microsoft to use their hardware is still paying. It’s just a different system of paying. From a cost/benefit analysis, this system of paying is better for many SMEs, but it’s not a one-size-fits-all holy grail.

Climatic Controls on Continental Environments, Fluvial Evolution and their Impacts on Human Societies in Holocene Asia
by Prof. Peter Clift

Indian mythology (the Rig Veda) makes references to a mighty river that flowed parallel to the Indus river, known as the Saraswati. No such river exists today, only the Thar Desert. Clift tries to find out if any of the current Indus river tributaries drained into the Saraswati in the past, by building mineral profiles of Indus’ different tributaries and comparing them with mineral profiles of sediments downstream. Since different tributaries eroded different regions of different geological ages, the sediment they contribute dates to different ages.

The Indus River today from http://geography.howstuffworks.com

The puzzle was that where the Saraswati once flowed through remains the river Ghaggar. Ghaggar is a weak little river and the tributaries that merge into it are unlikely to have provided most of Saraswati’s water. Another possible major tributary for Saraswati was the big river Sutlej, which now drains into the Indus river. However, the age profiles of the these tributaries would not be able to make up a composite that fits the sediments found where the Saraswati once flowed. There is one more river that can be added to make up the proper sediment profile: the big river Jumna, which now flows eastwards draining into the Ganges river. If the Jumna used to drain westwards, the amount of water it would have fed into Ghaggar plus the water from Sutlej would have made the Saraswati a huge river.

How could a river just flip directions? Rivers flow downhill so to flip direction the river would have to travel uphill. Jumna is an unusual river in that it lies right on of the high terrain between the Indus and Ganges basins. It was easier for Jumna to flip-flop between East and West. When Jumna flipped Eastwards, the amount of water in the Saraswati reduced. In that era, the monsoon was also weakening, so there was less water flowing through the rivers. This caused the Saraswati to slowly silt up and disappear.

I wonder if the Jumna might flip West in the future, causing the Saraswati to reform?

Clift also talked something about the Red River in Vietnam, but I didn’t get what he was saying about it enough to describe it here.

Fluid exchange flows: from magma chambers to volcanic craters
Prof. Herbert Huppert

Huppert talked about his research involving flow exchange between less dense fluid in a lower reservoir and denser fluid in a higher reservoir. The type of flow that takes place changes depending on the ratio of their density.

What I found fascinating was his experiment of water from a closed higher container falling into a closed lower container. Once a significant amount of air gathered in the upper container, the air flow upwards becomes “blurbby” — beads of air go upwards through the water periodically. And once a large amount of air had collected, air would jet into the upper chamber explosively. Some of the geologists present pointed out that this resembled the behaviour of some volcanoes.

The chapter “Geological Fluid Dynamics” in a textbook available on Huppert’s homepage. Here it is.

The other thing he talked about was what I think about as a “reverse gravity current” — only because I think of gravity currents as cold gravity currents. This is when a denser fluid enters a less dense fluid and undermines it. The everyday example that is unfortunately not very intuitive to me is that of opening the door of a heated house and having cold (winter) air rush in. Happily what was presented was the “reverse” situation when the door of a cooled building is opened — warm air rushes in, and keeps rushing in. And that is why we shouldn’t leave the door to air-conditioned buildings open. Unless there is one of those vertical air conditioners installed that is. My intuitive notion was that those things kept the cold air in, but I would say they actually keep the warm air out. Of course, it’s the same thing because if warm air can’t get in, the cold air doesn’t go out, but it’s not quite the same thing.

Knowns and Unknowns of the Climate Change Debate
by Dr. Adam Switzer

This talk didn’t turn out to be what I’d expected, but Switzer makes some good points for the audience in spelling out two things. And I learnt about a nifty thing that makes attending the talk worthwhile.

1) Then “known knowns” and the “known unknowns”

It is to Switzer’s credit that he not only lists down what is known about climate change, and what are unclear areas still under study. Often times, people trying to make a case of climate change do not do the second because they think it weakens their case. Clouds, solar effects, biosphere feedback, for example, are areas that still need a lot of study.

2) Lay people don’t the way scientists use language, so scientists should improve how they communicate

Now this point is worth a whole page by itself. However, I disagree with Switzer’s suggestion to use “fear” to motivate people, i.e. tell them what might happen if they do not act to reduce CO₂. My opinion is that scaring people is what government and gangsters do, not scientists. Fear will eventually cause a backlash — is already causing a backlash. Humans do not like to be afraid, so they will come up with all kind of reactions and excuses not to be afraid. When appealing to fear, we end up dealing with the illogical reactions to fear too.

Unfortunately, Switzer got some things wrong in the talk. Of course, he can’t be blamed because he is a geomorphologist. But these things are worth pointing out for people who have the same ideas.

The first point is that melting ice caps will not result in an increase in water vapour (a greenhouse gas) in the atmosphere. Melting ice caps will be concurrent with the increase of water vapour in the atmosphere, but not because the water in the ice caps become water vapour and goes into the atmosphere. The amount of water vapour is at equilibrium with other water states determined by temperature. When the temperature rises, the equilibrium state changes such that there will be more water vapour in the atmosphere, which will be concurrent with the ice caps melting but not the result of it.

The second point is the use of any local yearly temperature series to show global warming. In this case, it is Singapore, probably the Changi station, and the greatest contribution to the huge ground temperature rise is the urbanisation effect. Especially if it is a ground temperature measurement. If we demand climate change skeptics to be scientifically rigorous, the same standards must apply to the proponents as well. Local warming cannot be used to make a case for global warming.

Ok…now for the fascinating bit of information that made the talk worthwhile. Sea levels are rising, but one region where we would not expect sea level to fall is the East coast of North America. The reason, according to Switzer, according to another colleague of his who studies this thing, is that melting ice in the Arctic region “sucks” sea water towards it, causing an indentation. (In a hand-wavy way; I don’t pretend to understand what that means.) Now isn’t that interesting?

Go Blue to get Green – Roadmap to Building a Green Data Centre
Oh Bee Joo

Cheesy and stupid titles aside, this was quite an educational talk from the point of view of an operational guy (gal). Someone who is in charge of implementing the hardware and dealing with the logistics of a data centre. Oh may speak in Singlish, but she gives me the impression that she really knows her job. From ancient punchcard programming to plastic sheet-covered data computation centres by the roadside in developing countries, she says she has seen it all.

People often think of being environmentally friendly as a costly thing. I think this is because the real cost is insulated from them, whether because it is not their responsibility do the accountancy, or because of unseen subsidies on the national front. Oh and her employer IBM want a “green” data centre, or an energy efficient data centre, but they don’t want it out of the goodness of their hearts. Their clients don’t want it out of the goodness of their hearts. They want it out of the goodness of their pockets.

Oh discussed how changing very simple design features of the traditional centre resulted in greater efficiency in space and energy usage. Efficiency means less is needed, and less money/energy means lesser monetary cost. Melvin Soh, the centre manager who gave the talk after her, estimated the cost saved as 38K/year. So what are the main ideas for efficiency that I got from the talk?

1) It’s not James Bond.

Traditionally, people who run date centres are afraid that their maintenance guys, as in the people who deal with the air-conditioning and dehumidifiers, would sabotage their nodes. So a lot of space was wasted to seal off the support machinery from the computation machines. If we don’t assume that maintenance guys are out to sabotage the computers, and assume that they are not out to sabotage the computers instead, then there is no need to waste that space.

It kind of makes sense. If I were the maintenance guy who is going to be trackable when I enter the data centre, I will probably stick to my air-conditioner and not throw rocks at the computing racks.

2) Think small and modular

When I was a student trying to make pocket money by babysitting in the university computing centre, I found that the job was only cool in theory, but extremely uncool in reality. Maybe too cool in reality. This was because the room where the supercomputers and clusters were housed was kept really cold, the colder the better. I had to turtle in a small unventilated room that was insulated from the main data centre, and went out periodically to check that the thermometers. 13 degrees? Goooood.

So it was surprising when I found that Oh built a data centre to run at 25 degrees. The difference was the decision to cool the racks individually, rather than cool the entire room where the racks were placed. The nodes were designed to be flatter (smaller width) than they were traditionally, and placed in racks that were flatter. This made it easier for cooling fans to blow the hot air through the rack door, which was the cooling unit!

The rack door was covered with pipes through which cold water was piped. The cold water is circulated from a cooler/pump which I presume is somewhere else (not in the data centre, otherwise where does the heat go?). Temperatures are controlled to reduced condensation. Since each rack is cooled individually, there is no need to turn the entire room into a freezer.

Oh points out that the technology used has already been developed and used, such as for supercomputers. It was actually just changing the way people thought about the architecture of data centres. I think this is the main lesson to take away. Small, simple and cheap changes can result in big results.

I was at a GPGPU programming workshop this week. I am impressed! A 15 times or greater speed-up in the serial code I run on my personal computer, can turn the one month computation into a 2 day computation. Or I can triple the domain size and still do it in one week.

So what is GPGPU programming? It is General Purpose Graphics Processor Programming, or performing non-graphics computations on graphical cards by disguising those computations as graphical operations.

When we think of supercomputers, we usually think of some kind of parallel architecture. Many nodes perform computations simultaneously. Within each node are mutiple CPUs — I hear they’ve gotten up to 6 these days — so you split up the computation between nodes which split them up between CPUs, and since many CPUs are working on the computation simultaneously in parallel, you can do the job in a shorter time.

It is actually a bit more complicated than this, because communication between nodes can be substantial (and costs to reduce internode connection can be substantial), so changing the way a job is split up can make huge differences in computation time. This is as much as I know about this subject, which is basically not much.

Our personal computers are akin to one of the nodes. (In Beowulf clusters, the nodes are personal computers.) They usually come with 2 or 4 CPUs. But what if…what if the supercomputer architecture is already present, in a cheap form? What if development of the technology is driven not but cranky scientists but by the masses who want it for their entertainment and are willing to burn money for it?

The what ifs are already present in the form of graphics cards.

When I first heard of GPGPU programming, the notion was still relatively new and it was not feasible to get into it. To do calculations of interest it was necessary to formulate one’s codes as graphic computations and trick the graphics card into doing it. A few years later, the libraries to do so have appeared and are spreading.

My boss was not impressed. Supercomputers still give better performance, he pointed out, and models are written for supercomputers. Sure, they are. But I think he misses the point. Large, highly specialised models will always run on “supercomputers”. But nobody said what kind of architecture “supercomputers” should have. A “Supercomputer” is a “super” computer. What GPGPU programming does, is that is brings computation power on a single personal computer dramatically closer to current supercomputer standards, for a fraction of the price and manpower. (I’ve worked in a supercomputer facility before, and it takes manpower to keep those things running. The older ones tend to get really finicky.) A super-version of this would be … super-super.

People have realised this, and are developing GPU-supercomputers. This technology is going to take time to mature. But the computation power available on a single personal computer at low price is already available. It’s true that you can write parallel code to run on supercomputers. I have done that before, once, in a class, and it was a horribly painful process and the knowledge was swiftly purged out of my brain as soon as I finished the class. I am only passably competent in programming, so anything that’s too complicated goes over my head. In the workshop, I found the code needed to use the GPGPU library (CUDA) easy to make sense of. The way I see it, if it’s not too hard to use, and even if one spends a week on it, and another week to compute, it’s faster than spending four weeks waiting for one’s results to turn in. Especially since one will inevitably be forced to do recalculations.

So imagine a this kind of semi-supercomputing power available for $500 to all scientists involved in computation — the end-users, not the ones running specialised models on supercomputers. Certainly, even the ones who have written parallel code to run on supercomputers, but are jostling for computation time with other equally cranky users. Then think about clusters of GPU-computer, with multiple times that powers. How can this not be something to get excited about?

The Quest for Immortality
Singapore National Museum until 4 April 2010

There were two burning questions I wanted to ask about Egyptian mummies. The first was the rather morbid one of brain extraction. Mummies, we are told, have their brains extracted through the nostrils. Since nostrils are small holes and the brain a rather big mass, they had to suck that mass (mess) out somehow. How? Stick a reed in and suck it out?

The answer to that, according to the curator, was that they used a hook to pull the brain out. I don’t understand that. The other method, according to him, was the back of the head was cut opened and the brain drawn out. Fine, I can accept that.

The second question, a just as puerile one I guess, is what are the things stuck on their chins. Are those things beards? They look like toilet rolls, not beards, or maybe that’s as good as the sculptors could do in those times? Doubt it. They made pretty realistic ibises (picture on the right), so they should be able to make beards.

I never got to ask the question, but some googling turned out the answer: most Egyptians are clean-shaven, but the pharaoh uniform included a beard, which was a symbol of kingly power. The beard requirement did not deter the godly pharaohs from shaving; they just wore a fake beard! Even the female pharaohs wore fake beards.

If you plan to go to the exhibition, try to go for one of the guided tours, preferably during a non-peak period. The guide/curator was very amusing and full of tidbits of information. Listening to him talk about the exhibits beat reading the placards.

One thing that amazed me more than anything else about the exhibits was how some of the pigments were so well-kept that even after thousands of years, the colours remained as vibrant as any modern water colour! This is something that has to be seen to be believed.

The other thing that was mind-boggling was the elaborate extent to which the (rich) egyptians went to ensure their life after death. Families still have to bring the dead food and there are even emergency rations in the tomb if the family forgot to bring food. We might be tempted to laugh, but think about this: What beliefs about Afterlife do modern people have?

Orang Botak recommends The Quest for Immortality.

Vocal Recital by Zhao Yunhong
Piano accompaniment Li Bin

In the age of on-demand music on the internet, free clips on Youtube, and perfection preserved on music CDs, why go to a life performance?

Live performance still has its place in the world. Firstly, it is almost impossible to capture the actual sound quality of a live performance. There is something in the timbre of the sound passing through the air and hitting one’s eardrums, altered by the nature of the location itself, that is incomparable. Secondly, in a live performance, it is not perfection one seeks, but imperfection.

The fact that this recital took place in a small theatre and that I was less than 10 rows away from the performer made it doubly the treat. I could see every twitch on the singer’s face, and the movement of the pianist. (It’s amazing how he managed to flip the pages of the score while playing, by the way.)

I don’t think Zhao is that good. There were some parts which she sounded a bit off, especially when the performance wore on and she got tired. But that is the fun part. We have to keep in mind that this is not a once in a century soprano (as posted on youtube) or a machine generating sounds. Parts I enjoyed: “Rejoice Greatly” from Handel’s Messiah, “Je veux vivre” from Gounod’s Romeo and Juliet, and “Les oiseaux dans la charmille” from Offenbach’s Les Contes d’Hoffman.

EMD stands for Empirical mode decomposition. I’ve never heard of this before term before this week, when I attended a talk by Huang and other researchers working to develop method. The paper that started it all is “The empirical mode decomposition and the Hilbert spectrum for nonlinear and non -stationary time series analysis” by Huang, et al (1998).

That’s ten years ago! I am so backwards. And the paper is nearly a hundred pages long. I’m don’t know how I’m going to find time to read it. (Add to the rapidly expanding to-read list).

I’m so excited by this method, and I’m impressed by some of the developments presented by the researchers. Really impressed. Ok, I’m easily impressed. My colleague pointed out that he didn’t know why anyone would use EMD instead of EOF. I am not sure either. But it’s the entire cleverness of the thing that impresses me, dammit!

The original method is this: you have a time-series of some data. You take its maxima and interpolate a spline out if those. Then you take its minima and interpolate a spline of those. Then you take the mean of those splines, and subtract it from the signal. Repeat this until the envelope of your maxima and minima splines is nearly symmetrical (resulting in a mean spline of zero).

This decomposes your signal into a bunch of orthogonal modes. (I don’t know why they are orthogonal. Huang says that he did not know they were orthogonal when he came up with the method, but it was proven they were orthogonal later.)

Another amazing method by Wu was to add noise to noisy data, and then average all of it out to extract the signal.

I wish I had time to attend all the talks, but I don’t (because I have to do my work too).

The beauty in these methods, is that the underlying notions are a very simple and beautiful ones. They are also very physical ones. (I’m not one of those pure mathematician types who just like mathematically beautiful concepts, so those with a physical basis impress me more.) Because it is saying, something important, powerful and useful, does not need to be complicated. If you can capture the fundamental essence of the problem, the rest is tidying up, closing up.

This is not to say everything can be simple. Or that tidying and closing up is inferior work. It’s actually tidying and closing up that is the hard work. (i.e. Putting the idea into the mathematical framework and doing the proofs, which as what one of the speakers was doing. The hard work of putting it in solid foundations. But then mathematicians probably think that is fun part). But every time I see one of these brilliant ideas, I am just so impressed and excited.

Makes me feel stupid too. I feel more and more stupid as time passes. Ah well… we can’t all be genuises.

Complex Systems
Professor Constantino Tsallis

Tsallis Entropy is a new concept to me, and I find myself not entirely comfortable with it, largely because I know very little about it other than what was presented at the start of the talk. Talks, I think, serve mainly to introduce threads of interesting subjects to us. We have to follow the threads ourselves. I will do this when I have time, but here is a summary of the idea of Tsallis Entropy:

Boltzmann-Gibbs type statistical physics are derived from ideal-gas assumptions where there is no interaction between particles. Therefore, they suffice for the description of systems with no or short range interaction between particles. Statistical physics must be generalised to include systems with long range interactions, by introducing a factor q. In “normal” Boltzmann-Gibbs statistics, q = 1, but q does not have to be 1. If I interpret his response to one of the audience correctly, Tsallis believes that the factor q is a parameter describing of how quickly trajectories in phase space diverge when the system is just at the edge of chaos, but not yet chaotic yet.

An amazing empirical evidence of q-statistical physics is the existence of the q-triplet in the solar wind. I don’t understand what the q-triplet is, but it is supposed to be three stable states for a system following q-statistical physics. There was a discussion how the solar wind q-triplet values were arithmetic, geometric and harmonic means of each other. (Specifically, 1 subtract the q values did.) A mathematician in the audience suggested that this could be a mathematical result of the type of assumptions about the forms the q values took (q₁ = 2 – 1/q₂ …). This I thought quite possible so I went back and fooled around with algebra for some time, but I didn’t get the means result evident from the assumptions. I’m not a mathematician, so many that guy in the audience would follow up and find out why.

The main thing I wasn’t comfortable with was how a system can be extensive, but not additive. In classical thermodynamics we always considered the two to be the same. Ok – so maybe that’s why it’s “classical”, and this is “new”. Tsallis has published a textbook this year for the interested. Introduction to Nonextensive Statistical Mechanics: Approaching a Complex World.