Slashdot Mirror
The BBC's Distributed Climate Prediction
CongoJoe writes "The BBC has teamed up with Oxford University to conduct the world's most ambitious climate modeling experiment." From the article: "Trying to predict climate change is hard. There are lots of factors involved - air temperature, sea temperature and cloud cover all play a part - as do dozens of other variables. Therefore, there are a huge number of calculations involved ... Using a technique known as distributed computing, we're hoping to harness the power of thousands of PCs around the world. If 10,000 people sign up, we'll be faster than the world's biggest computer. And we're hoping to be even better than that."
if those 10,000 people turned their computers off, when not in use, instead?
Not that I don't think this is a good investment of spare cycles. I'm just wondering what the power savings would be, as an alternative.
Also, I notice there is no OSX client, only Windows and Linux.
I am concerned about sending my spare CPU time overseas, i mean, they say they are "climate modeling", but could they not just as easily be selling this "distributed" computer network to the black market, or Iran, or doing nuclear weapons simulations?
:)
I mean, they ARE british, and you all know what happened last time they got their hands on a bunch of CPU time- that poor kursk...
How bout the dental hygene simulator project?
sorry, just had to rip on some of my brit friends today
Case/PSU: Asus TA-211
M/B: Asus K8V-X SE
CPU: AMD Sempron 2800+
I suspect cpdn gets a lot more work out of each client. Rather than breaking the job into little work units like S@H, each client runs a complete simulation lasting hundreds of (simulated) years, and taking weeks of heavy cpu usage to complete. At least this was the case when I took part ages ago, it kept my cpu working very hard.
Oh no... it's the future.
So no, the BBC experiment itself doesn't have an OSX client.
Sulphur cycle models now take about 1200 hours on my 2.4 GHz X2* to complete. Trickles are sent about every 7-8 hours I think. So you get credit every 7-8 hours, and bits of the results are uploaded. You can follow along on the graphs they provide.
I'm running a Spinup test model, they want 200 simulated years. It's taken 1891 hours for 68.69%. You'll never get this model though, so you don't have to worry about this much computing time.
* One model per core
..that people applaud something like Mersenne's Prime or SETI when spare CPU cycles are concerned,
...
but when an action to the benefit of everyone - namely checking up on the environment - is undertaken using the same technology that those same people start commenting on artificial climate change:
"what about the power usage", "this will cause more rapid climate change" etc etc.
Is it either that
a. those people believe the climate WILL change through use of more computers
1. but do not want to know about it
2. and feel that any action towards proving it further is useless
b. are in the climate-does-not-change-artificially group and will take any reason - even those undermining their own stance - to support it because....
1. they do not want to know whether they are wrong
2. feel that there is enough proof already that the environment is not changing
Could anyone explain the logic behind the reasons A1-B2 ?
Slashdot: stuff for news, nerds that matter, matter for news, stuff that nerd
In a long term ClimatePrediction.net user let me answer a few questions:
1. the extra electricity used is insignificant in global terms. In winter the extra heat generated means you don't need to keep your central heating so high. Super-computers use lots of electricity themselves so running these same simulations on one (actually would take dozens) would not be better.
2. We are currently at just under 50,000 users, sounds a lot but there are millions of potential models to crunch - although useful science can be done with a smaller number. I would expect over 100,000 users to enrole in total. Some users are experiencing problems, they seem a very small percentage of the total judging by the posts on the help lines. Many of the problems are due to trying to run it one machines not powerful enough, climate simulations are heavy duty programs and need a beefy machine to run in a reasonable time.
3. The exeriment is an 80 year hindcast + 80 year forecast 1920-2080, even if a model does not complete the hindcast will be useful. The second BBC program is scheduled for May which is the "end" of the project, in pratice the scientists will be able to use results reported for at least a year after than. Its the science that counts.
4. The data is not being hoarded by the BBC, it is kept by the ClimatePrediction.net team, and is available to scientist throughout the world.
5. in a few days the ClimatePrediction.net servers will start dishing out TCM models to their users, that will add another 45,000 odd machines.
6. ClimatePrediction.net is in the midst of a sulphur cycle experiment, which compelements the 2x CO2 doubling and THC slowdown experients. There is at least one other experiment in beta.
7. ClimatePrediction.net has already had one major paper in Nature, as well as many others, this is the distributed project that seems to producing the best science.
If you don't trust the BBC, then sign up via the original web site at Oxford University. This distributed computing experiment as been running for a couple years (I was a beta tester). They have already published some initial results.
I just downloaded and installed it on FC4, via their shell script, and it has a GUI that works just fine.
What's your GCNSEQNO?