climateprediction.net home page
Project Losing it's Way?

Project Losing it's Way?

Message boards : Number crunching : Project Losing it's Way?
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile JIM

Send message
Joined: 31 Dec 07
Posts: 1086
Credit: 19,554,876
RAC: 805
Message 59851 - Posted: 21 Mar 2019, 4:30:42 UTC

I have been running CPDN for a long time. I go back to the BBC experiment about 12 years ago. I am starting to worry that this project is about to lose the thing that made it and other distributed computing project a success in the fist place. That thing is models that run on average home computers. Recently, there has been a lot of talk about new super-high-resolution models that may not be practical to run on average home computers.

Due to cost I am never likely to have a 3.5+ GHz machine with 12 Gb’s of RAM. There are no Ryzen 2700’s processors in my future. My 3 computers are mid-level laptops. Clock speed is about 2.6 GHz with 8 Gb’s of RAM.

What worries me is that the resource requirements will grow until average home computers, mine included, will not be able to finish a model in a reasonable amount of time.

In the old days we were told that regardless of how long it took to finish a model the results would be excepted and used by the researchers. But, that was in the days when they were developing the models. Now that they are in the production phase the researchers are less patient. They want their result quickly.

I don’t want to invest 6 or 8 month running a WU only to have the researchers loose patience and close the batch and reissue it to someone else with a faster machine.

I just hope that after all this time I am not about to be frozen out.
ID: 59851 · Report as offensive     Reply Quote
Profile geophi
Volunteer moderator

Send message
Joined: 7 Aug 04
Posts: 1821
Credit: 36,393,036
RAC: 10
Message 59852 - Posted: 21 Mar 2019, 5:06:17 UTC

I can't completely foresee what requirements will be 1 or 3 or more years into the future. But there's certainly some things you could do to turn the models around quicker when and if it becomes a problem. With the CPUs you have, they are all dual core with hyperthreading, so they look like quad-cores to boinc. Running at most two models at a time instead of four, will certainly help the speed. It might not only help the speed due to not using the hyperthreading threads which aren't real cores, but also because the temps could be lower and therefore thermal throttling less of an issue, i.e. higher average CPU core speeds. Maybe, for example, your 2450M, with decent airflow around it, would run the cores at a higher turbo boost instead of at lower speeds. I'm just guessing since I don't know what speed the cores on your laptop are running at when fully loaded.

But on the cpdn side, I hope they create different models names, even if the only difference is the name, and associate one model name with the higher res and/or longer running models, and the other model name with the shorter or lower-res models. That would still provide work, hopefully useful, for the computers that aren't all that speedy.
ID: 59852 · Report as offensive     Reply Quote
Profile Dave Jackson
Volunteer moderator

Send message
Joined: 15 May 09
Posts: 2463
Credit: 3,124,201
RAC: 383
Message 59853 - Posted: 21 Mar 2019, 6:19:25 UTC - in response to Message 59852.  

Running at most two models at a time instead of four, will certainly help the speed.


I too have been in the game since the BBC experiments though initially with a different user ID that got lost somewhere along the way. I also have two slow computers. a dual core 2.7GHz desktop and a 2.16GHz laptop with 4 and 8GB ram respectively. Some of the recent models do run faster on the laptop if I only run two or three. (no hyperthreading here either. I suspect this is due to an memory i/o bottleneck of somewhere. I should really try and get some figures but at the moment I am not running all four cores to try and get some of the more fragile tasks through without seg faulting.

I agree with George about the names which would enable some tasks to only go to the faster machines. I am also unlikely to be getting a significantly faster machine in the short term at least though at some state I do intend to do a mother board and cpu transplant on the desktop.
ID: 59853 · Report as offensive     Reply Quote
Jim1348

Send message
Joined: 15 Jan 06
Posts: 372
Credit: 15,130,852
RAC: 56
Message 59854 - Posted: 21 Mar 2019, 7:15:26 UTC

I usually leave projects these days because they don't have enough work, and don't make enough use of my machines. I always have plenty of memory (minimum 8 GB and usually 16 or 32 GB).

I think I was one of the first to use large ramdisks with all that memory years ago in order to reduce the error rates, which were often due to the disks in those days not being able to keep up with the high write rates. I am even willing to change between Linux and Windows to accommodate them. The only real question is: are they doing anything worthwhile?

My main concern about CPDN is that it is just spinning numbers for the case of academic studies of past weather events. If they want to keep me, they will make the studies more relevant to the real world.
ID: 59854 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 6952
Credit: 20,843,205
RAC: 0
Message 59855 - Posted: 21 Mar 2019, 8:03:15 UTC - in response to Message 59854.  

Well, studying past weather events can teach them about how the climate may change in various areas if humans do certain things.

And real world studies are most likely done by weather bureaus around the world.

It's probably all in the research papers that get published. The few that I've looked at are way beyond me in what they discuss, but no doubt quite meaningful for lots of others.
ID: 59855 · Report as offensive     Reply Quote
Profile tullio

Send message
Joined: 6 Aug 04
Posts: 264
Credit: 844,488
RAC: 0
Message 60003 - Posted: 24 Apr 2019, 6:25:26 UTC

I have a AMD A10-6700 CPU on a Windows PC, which also hosts a Linux Virtual Machine with SuSE Thimblewood, a development version with frequent upgrades. The PC had Windows 10, but a crash due to a bad RAM (it had 24 GB, now reduced to 20) made me use a repair disk which was made in Windows 8. So it is now running Windows 8, upgraded to 8.1. The funny thing is that when running Windows 10 it could run only two tasks at the same time, plus a GPU task from SETI, Einstein or GPUGrid, and on Windows 8.1 it can run 4 tasks. The CPU was sold as 4 cores, but the Windows Task Manager says it has two cores and 4 logical processors.It runs at a maximum of 3.70 GHz but I have seen it reaching 4.14 GHz, a speed that the Ryzen 5 1400 on another Windows 10 PC cannot reach. It is now running 2 SAM tasks, while two crashed due to frequent stops and starts in attempts to bring it back to Windows 10.
Tullio
ID: 60003 · Report as offensive     Reply Quote
Iceberg

Send message
Joined: 28 Dec 17
Posts: 10
Credit: 766,872
RAC: 18
Message 60005 - Posted: 24 Apr 2019, 15:51:29 UTC - in response to Message 59855.  

I completely agree, Les. The way CPDN has evolved seems to be the way not only academia studies extreme weather and climate events, but how weather offices do as well. By developing a range of possible and probable outcomes as a result of various emissions scenarios (RCP25, RCP45, RCP85, etc.), it gives scientists, forecasters, and policy-makers a better understanding of what society is to face. Academic studies are frequently examined and written about in the popular media, too, if the media are doing their jobs properly, in a way the public can understand. Sadly, many media bureaus have drastically cut their science and environment desks, which is critically damaging to society, in part because false news gains traction without having a bulwark to refute it.
ID: 60005 · Report as offensive     Reply Quote
Profile tullio

Send message
Joined: 6 Aug 04
Posts: 264
Credit: 844,488
RAC: 0
Message 60009 - Posted: 25 Apr 2019, 11:08:21 UTC

Antonello Pasini is an Italian climate scientist. He hes met and spoken with Greta Thunberg and given her a copy of his book "From observations to simulations. A conceptual introduction to weather and climate modeling", World Scientific, Singapore.
He is running a blog on Climate science in "Le Scienze", the Italian edition of Scientific American, and is skilled in explaining complex problems. His background is in theoretical physics.
Tullio
ID: 60009 · Report as offensive     Reply Quote
Dave Roberts

Send message
Joined: 15 Jan 11
Posts: 165
Credit: 5,211,097
RAC: 57
Message 60012 - Posted: 25 Apr 2019, 14:26:33 UTC - in response to Message 59854.  
Last modified: 25 Apr 2019, 14:33:27 UTC

My main concern about CPDN is that it is just spinning numbers for the case of academic studies of past weather events. If they want to keep me, they will make the studies more relevant to the real world.


Just noticed this comment by Jim1348 and hope the following might be useful.

Running simulations over past times is absolutely critical in insuring that the models are as theoretically sound as possible when using them to simulate the future. They really are relevant to the real world

From the empirical data that we have acquired about past climate events and weather, we can propose a theoretical model of the worlds (area) climate which we can then attempt to verify by running it from a known starting point in the past to a known finishing point sometime later.

We can then check to see how closely the results from a run of the model match the known data.

The closer the model results are to the actual data, the more confident we can be that the model is a fair representation of that reality and will simulate the future reasonably accurately.

We often have to have a series of simulations of the past, refining the model both in terms of the theory and its parameters until its results are within reasonable and predefined probabilistic limits of (un)certainty.

It's from this point that we can play around with the model, both theory and parameters, to test the effects of human activity on the future climate.

It's a never ending challenge and generally speaking, every run is useful, even if it 'fails'.
ID: 60012 · Report as offensive     Reply Quote
Sarah Sparrow
Volunteer moderator
Project administrator

Send message
Joined: 30 Sep 15
Posts: 8
Credit: 91,760
RAC: 0
Message 60019 - Posted: 26 Apr 2019, 19:35:31 UTC - in response to Message 59851.  

Just to respond to some of your concerns and hopefully reassure you that as a project we are still heading in the right direction.

Although we are looking at running some higher resolution models our aim with these is that they should still be suitable to run on an average computer. We are aware that otherwise we will not be able to do the analysis that we wish to. One of the initial tests we do when considering new models is to check resource requirements and we intend to keep the bulk of work units within reasonable constraints otherwise we will reject bringing the model within the project. There are a couple of projects coming out that are of higher resolution (but still run ok on our test machines which are setup to represent an "average" system). These will be a minority of the runs that we send and will be sent out as a new distinct app so that people can choose whether to run these or not. In many cases initially they will be limited to a single OS as well.

We do gratefully receive all contributions of simulations that are run. However if we have enough simulations in place to do the analysis that we need to do then it seems more efficient to use peoples computing time on simulations that are needed so we close batches. This also allows us to do essential cleanup processes on our servers here. Generally batches are not closed until we have around 80% of the simulations successfully returned to us. We aim to send work units that do not take too long to run, although realise that in many cases these simulations may still take several weeks.

As for the analysis that we do, we do look at past events as part of validation but in many of the analyses we include possible future likelihoods of extreme events to make these relevant to policy and decision makers. Increasingly the studies we do are focussed not just on the extreme event that occurs (heatwaves, floods, droughts etc) but on also on the extended impacts (e.g. public health, infrastructure and economics), and likely change in risk of these, making them "more relevant to the real world". A lot of these are still in the pipeline, but you should hopefully see some of these appearing in the future on the publications page.
ID: 60019 · Report as offensive     Reply Quote
Jean-David Beyer

Send message
Joined: 5 Aug 04
Posts: 290
Credit: 3,075,733
RAC: 13
Message 60020 - Posted: 26 Apr 2019, 20:00:08 UTC - in response to Message 60019.  

There are a couple of projects coming out that are of higher resolution (but still run ok on our test machines which are setup to represent an "average" system). These will be a minority of the runs that we send and will be sent out as a new distinct app so that people can choose whether to run these or not. In many cases initially they will be limited to a single OS as well.


WCG have an option the users can select whereby they agree to accept beta testing versions of applications (or, more likely, potential applications). I see no reason why CPDN could not do the same.
ID: 60020 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 6952
Credit: 20,843,205
RAC: 0
Message 60021 - Posted: 26 Apr 2019, 20:59:36 UTC - in response to Message 60020.  

WCG have an option the users can select whereby they agree to accept beta testing versions of applications (or, more likely, potential applications). I see no reason why CPDN could not do the same.


Beta testing is strictly researchers and a few select experienced users.
And it doesn't take many computers to run 2-3 models.
ID: 60021 · Report as offensive     Reply Quote
Profile astroWX
Volunteer moderator

Send message
Joined: 5 Aug 04
Posts: 1496
Credit: 92,841,900
RAC: 68
Message 60022 - Posted: 26 Apr 2019, 21:02:46 UTC - in response to Message 60020.  

WCG have an option the users can select whereby they agree to accept beta testing versions of applications (or, more likely, potential applications). I see no reason why CPDN could not do the same.


WCG sends an occasional Beta task to my boxes. The WCG Beta demands tend to be significantly lesser than CPDN's tests.

There was a time the CPDN test site was opened to all. Too many 'set and forget' users joined, diminishing the effectiveness of the tests.

A separate CPDN test site was established. Somehow the URL leaked and spread. Not good. That site was closed and another established, with membership by invitation only. Regrets if that seems arrogant or elitist; it isn't meant to. It it intended to maintain the effectiveness of the development process.

I was part of the test environment for years. I refused the latest invitation, not because experience is lacking but because age made me less capable than in years gone by. Why? Because I believe in the mission's quality.

Thanks for your suggestion, Jean-David.
"We have met the enemy and he is us." -- Pogo
Greetings from coastal Washington state, the scenic US Pacific Northwest.
ID: 60022 · Report as offensive     Reply Quote

Message boards : Number crunching : Project Losing it's Way?

©2019 climateprediction.net