After writing the last blog I started looking out datasets to do some calculations along similar lines.
I wanted data to locate accurate figures for the growth in the number of computers in the world for the period 1990 - 2010. In particular figures for the growth of the desktop, laptop, server, enterprise and HPC markets. For the embedded market there exists reasonable data which is widely accepted.
You'd think that it would be easy to find something like that. Not so it appears. So I then looked at proxies for the data. Similar issues there: in many cases worse.
Finding information that is credible has proved to be much more difficult than I supposed. Those datasets that do exist are provided by, for example, hardware manufacturers or by suppliers of software and open to reasonable challenge as they would naturally support their perspective.
So here's the plea. Does anyone have or know of a reliable set of data for the areas above? It must be easily accessible and in the public domain.
Either email me directly or comment as below.
Sunday, 26 September 2010
Thursday, 2 September 2010
Energy efficiency - are we all barking up the wrong tree?
So... the holidays are over. Shame.
I got back to an interesting article in The Economist1 itself quoting an article in Journal of Applied Physics: D. A long time ago I used to contribute many moons ago to J. Appl Phys C on occasion, and so was fascinated to discover what on earth The Economist could have found in such a recondite journal. And it turned out to be very interesting. The headline is that by reducing power consumption of individual devices (in this case lights) we are simply creating greater usage of power.
In the article Tsao et al.2, predict that the introduction of solid state lighting could increase the consumption of light by a factor of ten within twenty years. They assumed that within the next twenty years the efficiency of lighting will increase by a factor of three, and that the price of energy wuill remain approximately constant in real terms. Under these assumptions then the number of megalumen-hours (a unit of the measure of light as perceived by the human eye) consumed by the average person will rise from 20 to 202 megalumen-hours. For a comparison (according to sources cited) in 1700 the average Brit consumed (generated?) 580 lumen-hours in the course of a year. Today that figure is 46 megalumen hours, an increase of a factor of about 80,000. The problem is that as long as people want to light areas ever more brightly we consume ever more electricity. And there is no sign that this is going to decrease. To compare recall that when people started using compact low-voltage lamps they greatly increased the number of lights and didn't simply stay with the same level of lighting. In many cases lighting levels measure in terms of wattage consumed went up by a factor of four to ten. In other words there is a competition between decreasing cost of lighting (as measured in cost per lumen-hour) and the increase in the desire to light areas (as measured by overall lighting consumption).
Tsao et al argue that the only way that overall lighting levels might drop off would be if electricity prices trebled and then not for twenty years. There are of course other controls that can be placed on excessive use of lighting; in the UK councils are starting to subject lighting levels, particularly external ones, to planning controls.
So what does this have to do with multicore? A lot actually. Both articles (The Economist's précis and the original )set me thinking about the computer evolution and the growth of multicore. Over the past forty to fifty years the computer industry has gone from being a (relative) minnow to a booming giant. With it has gone rapid increase in energy consumption and the Total Cost of Ownership (TCO) of compute systems. TCO and in particular overall power costs were a key driver in the industry's goal of reducing power consumption and increasing efficiency. This lead in part to the evolution of multicore. However, at the same time we place ever greater demands on systems. On the one hand we have a decrease in thermal output and power consumption and on the other hand we have continuously growing demands. Increased efficiency has given us increased headroom within existing power consumption envelopes. But for how long can - or will - that remain true? While we may be increasing efficiency, it may not continue to be possible for us to keep ahead of overall demand and there is some limited evidence that this is so3.
If we can't what do we do next? It seems to me that the only thing that we can actually do is to make more efficient use of the computer power that we have. There is a measure which corresponds roughly to program output (throughput?) per watt consumed, which needs to be reduced faster than the increase in demand.
1 The Economist, August 28th 2010 p66
2 J Y Tsao, H D Saunders , J R Creighton, M E Coltrin and J A Simmons, "Solid-state lighting: an energy-economics perspective" J Phys D: Appl. Phys. 43 354001
3 see also my eearlier blog "CO2 and Total Cost of Ownership"
I got back to an interesting article in The Economist1 itself quoting an article in Journal of Applied Physics: D. A long time ago I used to contribute many moons ago to J. Appl Phys C on occasion, and so was fascinated to discover what on earth The Economist could have found in such a recondite journal. And it turned out to be very interesting. The headline is that by reducing power consumption of individual devices (in this case lights) we are simply creating greater usage of power.
In the article Tsao et al.2, predict that the introduction of solid state lighting could increase the consumption of light by a factor of ten within twenty years. They assumed that within the next twenty years the efficiency of lighting will increase by a factor of three, and that the price of energy wuill remain approximately constant in real terms. Under these assumptions then the number of megalumen-hours (a unit of the measure of light as perceived by the human eye) consumed by the average person will rise from 20 to 202 megalumen-hours. For a comparison (according to sources cited) in 1700 the average Brit consumed (generated?) 580 lumen-hours in the course of a year. Today that figure is 46 megalumen hours, an increase of a factor of about 80,000. The problem is that as long as people want to light areas ever more brightly we consume ever more electricity. And there is no sign that this is going to decrease. To compare recall that when people started using compact low-voltage lamps they greatly increased the number of lights and didn't simply stay with the same level of lighting. In many cases lighting levels measure in terms of wattage consumed went up by a factor of four to ten. In other words there is a competition between decreasing cost of lighting (as measured in cost per lumen-hour) and the increase in the desire to light areas (as measured by overall lighting consumption).
Tsao et al argue that the only way that overall lighting levels might drop off would be if electricity prices trebled and then not for twenty years. There are of course other controls that can be placed on excessive use of lighting; in the UK councils are starting to subject lighting levels, particularly external ones, to planning controls.
So what does this have to do with multicore? A lot actually. Both articles (The Economist's précis and the original )set me thinking about the computer evolution and the growth of multicore. Over the past forty to fifty years the computer industry has gone from being a (relative) minnow to a booming giant. With it has gone rapid increase in energy consumption and the Total Cost of Ownership (TCO) of compute systems. TCO and in particular overall power costs were a key driver in the industry's goal of reducing power consumption and increasing efficiency. This lead in part to the evolution of multicore. However, at the same time we place ever greater demands on systems. On the one hand we have a decrease in thermal output and power consumption and on the other hand we have continuously growing demands. Increased efficiency has given us increased headroom within existing power consumption envelopes. But for how long can - or will - that remain true? While we may be increasing efficiency, it may not continue to be possible for us to keep ahead of overall demand and there is some limited evidence that this is so3.
If we can't what do we do next? It seems to me that the only thing that we can actually do is to make more efficient use of the computer power that we have. There is a measure which corresponds roughly to program output (throughput?) per watt consumed, which needs to be reduced faster than the increase in demand.
1 The Economist, August 28th 2010 p66
2 J Y Tsao, H D Saunders , J R Creighton, M E Coltrin and J A Simmons, "Solid-state lighting: an energy-economics perspective" J Phys D: Appl. Phys. 43 354001
3 see also my eearlier blog "CO2 and Total Cost of Ownership"
Subscribe to:
Posts (Atom)
Posts
