Amazing chart on immigration/emigration of researchers

I don't think that much needs to be said about the picture, but here you go anyway..

- Other than Belgium and Denmark (where the US is barely #2), the US is the #1 choice for emigration
- I'd assume China's emigration fraction will be similar to India's, but the striking thing about Indian emigres is that they (we) pretty much flock only to the US, whereas a decent number of Chinese also move to Canada
- Man.. look at those Germans move around !

Linky http://spectrum.ieee.org/at-work/tech-careers/the-global-brain-trade#.UE_uhs1ViPU.facebook

Cluster computing on the cloud (by the numbers)

I was recently involved in some consulting work with a small aircraft company and had to look up computing costs on privately owned clouds. I've always pooh-poohed this in the past because it seemed too expensive and cumbersome. The numbers have, however, gotten to a point where a second look is required. The following numbers are for the Amazon cloud and the rest of the vendors are competitive with this:

Usage of 16 cores costs $2.4 / hour and the first GB of data transfer is free.

If you're a bit more serious and are thinking long term, a yearly lease of 16 cores costs $4148 (up front) + $ 0.54 / hr and $0.12/GB. These costs can go down if you're thinking even more big time.

So let's crunch some numbers here... Assuming we need 16 cores at 60% utility over an entire year, the cost comes to $7k + data transfer. For a 3 year lease, it is $15k + data transfer.

96 such cores over 3 years thus cost $90k + data transfer. Unless you're running calcs that you shouldn't be running, data transfer won't be more than $10k, so we're looking at $100k over 3 years.

Compare this with your own 96 core cluster that probably is really useful for only 3 years - It costs $25k not including cooling, power, space and system administration. The last two items are definitely non-trivial. At a place like Stanford (or anywhere else, for that matter) you have to move heaven and earth to carve out space and a system admin costs $200k/year (including benefits - So even assuming that maintaining your cluster constitutes less than 5% of his/her responsibility, we're looking at non-trivial numbers). All said and done, it won't be unfair to put the real cost at $50k over 3 years.

Factor of 2 ? wow! Even if you just consider the cost of your cluster, it is just a factor of 4.

Clouds are here to stay... even from a HPC perspective

A Sense of Scale

Apple's recent announcement of it's cash pile and the California government's much maligned money problems got me thinking about the following numbers:

California:

Population: 39 Million
GDP : $ 2 Trillion
Budget deficit: $10 Billion
Total debt: $ 360 Billion

Apple Inc:

Employees: 60 K
Revenue: $ 110 Billion
Profit : $ 26 Billion
Cash Reserves: $ 100 Billion

Stanford University:

Staff+Students: 20 K
Revenue: $3.8 Billion
Net Inflow: $4.0 Billion
Total Endowment: $ 18 Billion

 

To be continued....

Energy use and GDP

Per capita electricity use / day
USA : 40 units (of which 36% is residential)
EU : 20 units (35% is residential)
China : 7 units (15% is residential
India : 2.5 units (!) (20% is residential)
World : 8 units (27% is residential)

Take a look at this very interesting graph..  
A good straight-line fit gives us a GDP of 40 cents for every unit of energy (total, not just electricity) consumed.  I know there are some implicit, non causal relationships, but this is a wonderful number to remember.. say for India to reach a per capita GDP of $10000, the annual per capita energy use could be close to 25000 units. Another way of looking at it (a Nobel winning Stanford Prof says this): For (roughly) every dollar spent on energy, you typically generate 4 dollars of GDP.  Quite remarkably, this number is more or less the same across all type of economies including for the big bad US of A. Americans account for 25% of the world's energy consumption and account for 29% of its GDP. In other words, there are no shortcuts, at least so far.


India better be building one nuclear reactor every month for the next 20 years ?

Daily energy use of the world : 360 billion units
Global power consumption : 15 Tera watts
Guy with a plan:  http://www.stanford.edu/group/efmh/jacobson/Articles/I/susenergy2030.html

Some numbers related to Energy

Energy Densities (Amount of energy released when the given substance completely combusts with oxygen at standard atmospheric conditions)

Coal :          10-24 MJ/kg
Wood :        14 MJ/kg

Gasoline*:    44.2 MJ/kg (31.7 MJ/litre)
Natural gas: 53.9 MJ/kg (0.038 MJ/litre)
LNG          : 53.9 MJ/kg (22 MJ/litre)
Propane     : 50 MJ/kg (25 MJ/litre)
Biogas       : 28 MJ/kg  (7 MJ/litre)
Hydrogen : 119 MJ/kg (8.1 MJ/litre)

Lead-acid batteries : 0.125 MJ/kg
Li-Ion batteries : 0.432 MJ/kg
Alkaline batteries : 0.594 MJ/kg

* Diesel and Kerosene are within 3% of this number

Let's switch to convenient energy units i.e.,  1 unit = 1 kilo watt - hour = 3.6 MJ

So a kg of coal contains roughly 6 units of energy and a litre of Gasoline contains about 8.75 units of energy.

Note that the energy density of the best possible battery is about 100 times less than that of an equivalent mass of gasoline. Of course, electrical energy conversion is much more efficient than thermal energy conversion, but this fact is only going to buy us a factor of 2. In addition to being heavy,  battery electrodes are quite expensive as well. Even though things are improving, you should get a sense of what electric cars are up against.

If you assume that gasoline costs $3.50 / gallon, the price you are paying per unit of energy content comes to 11 cents. This is quite interesting because we pay around 10 cents / unit of residential electricity.

Cost / unit of energy content  (at wholesale prices) :

Coal  :  0.4 cents / unit
Natural gas : 1.5 cents / unit
Crude oil : 5 cents / unit

Note that so far, we've been just talking about energy content in fuels and haven't considered conversion losses. If we take everything into account and go by the wikipedia definition:

Levelized Energy Cost (LEC) is an economic assessment of the cost of the energy-generating system including all the costs over its lifetime: initial investment, operations and maintenance, cost of fuel and cost of capital,  this is what we get

 The chart shows cost / Mega watt - hr, so divide the last column (which is in dollars) by 1000 to get cost / unit of electricity. So according to this chart, conventional sources cost less than 10 cents / unit and land-based wind energy is competitive. Solar energy continues to be off the charts expensive even with all the recent advances.

Coal and natural gas based energy is dirt cheap and the big 3 energy consumers (USA, China and India) have enough reserves to last out this century at least as far as electricity use is concerned. Transportation fuel may be a bit more complicated with the approach of peak oil (and such), but I think gas to liquid technologies will come to the fore. Note that if more stringent emission restrictions are brought in on fossil fuels, wind energy has a good chance to be very viable and Solar can become borderline competitive.

Overall,  the cost of energy continues to be amazingly low considering the fact that the best cyclist in the planet can only give you 0.3 units of energy over an hour and animal labor isn't that much different. Yes! I compared the present situation with the stone age to make you feel better!

10 Myths about India (International edition): Call centers are very important to India's growth

When the topic of India gets brought up in chatting with my colleagues (and students) or friends (especially the long drawn discussions at the dog park), I frequently start off with "No, in reality, it is not like that.." So I thought I should put some of my points in one place (not in any particular order) to decrease entropy generation:

Myth #2: Call centers are very important to India's growth: Not really!. Out of India's $ 76 billion information technology (IT) industry, services such as call centers, tech support and medical transcription constitute a small fraction of the revenue. India has a healthy and growing IT industry that offers high quality software and consulting services on par with the best in the world. Call centers, on the other hand, account for about 1% of GDP and about 0.1% of the workforce, and since this sector does not involve a high degree of technical skill (I am appreciative of good communication skills) competition from other low cost markets (such as the Philippines) is growing. The state of the IT industry, overall, is better because it is not just dependent on low cost services.  The IT industry has changed some business approaches and social practices and has been very beneficial, but it still measures only up to 7% of  GDP and only recently have some companies started to focus on creating original, innovative products rather than rely on services. In comparison, the engineering industry (including mining and assorted stuff) contributes about 20-25% of the GDP (by some measures) and though traditional industries aren't as "world class" as IT services, specific sectors like automotives are getting there in patches.

10 Myths about India (International edition) : The education system in India should be really good

When the topic of India gets brought up in chatting with my colleagues (and students) or friends (especially the long drawn discussions at the dog park), I frequently start off with "No, in reality, it is not like that.." So I thought I should put some of my points in one place (not in any particular order) to decrease entropy generation:

Myth #1: The education system in India should be really good : You're very likely to hear this in the US and outside of the Indian elite, probably only in the US. My take on this myth might be good enough even for the Indian edition. While literacy rates in India have increased at an appreciable rate (from 12% when the British left us to about 76% currently and I am indeed proud of this), serious issues remain. The definition of basic literacy is downright depressing. A good fraction of primary and secondary schools are run by just one teacher (simultaneously teaching many classes while being absent half the time) and despite all the efforts, only 16% of students reach high school and only 8% graduate. This is absymally low for a rapidly growing economy with an acute shortage of skilled labor. It is true, however, that the fortunate few millions that end up in good private schools receive better training in mathematics and science than the average high school student in the US. So we have that going for us.

About 6% (of the total college age population) reach college and on average, the quality of education and employability of graduates continues to be very poor, and by all measures, is declining by the day. Adam Smith (in my book, a top 10 human being of all time) once said "…if an equal proportion of people were educated at the public expence, the competition would soon be so great, as to sink very much their pecuniary reward.” While this problem could be true in advanced economies (present day Britain, especially), this is not at all India's concern. In India, the issue is that with growth in the higher education sector, quality has taken a hit beyond imagination. Take the example of the southern state of Tamil Nadu - This is one of the most progressive and prosperous states in India (I would rate them in the Top 3 by both measures), engineering colleges have mushroomed in the last 10 years.  About 20 years ago, one in 80 aspirants could get into engineering colleges.. and these days about 20% (50000) of seats remain vacant. With the explosion in opportunities has arisen a tremendous shortfall in lecturers (quality and quantity wise) and while there are some very good teachers around, a good majority (not all, bless those souls) of lecturer posts get taken up by fresh college graduates and many of them do so because they can't find other decent options. Some reports that I have read suggest that companies think that less than one in 10 engineering graduates are "employable". Access to education is one of the most important aspects of a growing economy and while the increase in quantity has done short term good, the alarming decline in quality doesn't bode well for the future.

Secondary education is about quantity and Higher education is about quality. At the present moment, India is well behind the curve in both aspects. Changes can't and won't happen overnight, but I think things will get worse before getting better.

All this said, at the very top (IITs), the quality of education is still very good (but declining according to various sources) and comparable to the Ivy leagues and Oxbridge (in technical rigor, but much less well rounded). However, less than 1 in 200 aspirants make it to the original 6 IITs) and just the cream in many other universities is comparable. Every year, more than 100,000 Indians leave for higher education to the US (an equal number to the rest of the world) and a majority of these folks rely more on hard work, survival instincts and adaptability than background knowledge to make it to solid jobs, all the way to the top of the middle class. In fact, the average household income of Indians in the US is the highest among any group and is twice that of the US average. Of course, at the highest end of this spectrum, the best of the best Indians have made amazing contributions in founding big corporations and produced significant achievements in science and engineering.

In summary, the Indian education system has always faced incredible challenges (even at the present moment, only at par with the likes of Pakistan and Bangladesh on average) and is exhibiting growing pains. You probably have to be aware of the fact that it is indeed the top 5% of the crop that comes to the US. To boot, a majority of this group is successful not because of what they learnt in the class room, but rather off of it. I am, however, confident that with the growing awareness of India's acute problems and increasing pressure on the government to perform, things will improve in a few years.