[climateprediction.net] New study: STORMS: Investigating how low-pressure systems may change in the future

News and Information related to Distributed Computing
Post Reply
BOINC_News
Reactions:
Posts: 997
Joined: Sun Nov 08, 2020 3:51 pm

[climateprediction.net] New study: STORMS: Investigating how low-pressure systems may change in the future

Post by BOINC_News »

Title: Quantifying controls on the intensity, variability and impacts of extreme European STORMS Victoria Sinclair, Clément Bouvier
Institute for Atmospheric and Earth System Research,
University of Helsinki, Finland Throughout the year, low-pressure systems regularly move across Europe, usually from west to east, bringing cloud, rain and windy weather. Sometimes these weather systems can become very intense, and the winds and rain associated with them can cause damage to buildings and infrastructure, flooding, and can disrupt electricity supply and travel. Although the short-term weather forecasts of these storms are now quite accurate, it still remains uncertain how these storms, and their impacts, are likely to change in the future as our climate changes. Some of this uncertainty is because our understanding of what controls the strength and impacts of these storms is incomplete. The aim of this project is to understand what controls the strength and structure of these low-pressure systems. We will quantify how the atmospheric state that the low-pressure systems develop in affects the strength and structure of these low-pressure systems. This atmospheric state can be described by various parameters, for example, the mean temperature, moisture content, and upper-level wind speeds (i.e. the strength and width of the jet stream). Since there are lots of different parameters we want to study (not just the ones described above), we want to do lots of experiments in a high controlled manner. Therefore, we will run a large ensemble of simulations of idealised low-pressure systems using the numerical weather prediction model OpenIFS. Although the simulations are idealised, the weather systems that develop look very like real weather systems that we observed in reality. Each ensemble member differs in its initial atmospheric state, and we choose these initial states to cover everything from the current climate to past pre-industrial climates to the most extreme future climate projections. This is exciting because although idealised simulations of low-pressure systems have been performed before, this is the first time that such an extensive exploration of the parameter space will be conducted. Once we have the results from the large ensemble, we will calculate different measures of the strength of the storms and then use machine learning techniques to see how these relate to the initial states. Our results will hopefully increase in confidence in how these storms and their impacts will change in the future.

Source: https://www.climateprediction.net/new-s ... he-future/
Skillz
Site Admin
Reactions:
Posts: 1854
Joined: Sun Sep 15, 2019 3:03 pm

Re: [climateprediction.net] New study: STORMS: Investigating how low-pressure systems may change in the future

Post by Skillz »

Anything I need to know about running this project? Last time I remembered running this project the tasks took a really long time to complete.
StefanR5R
TAAT Member
Reactions:
Posts: 1661
Joined: Wed Sep 25, 2019 4:32 pm

Re: [climateprediction.net] New study: STORMS: Investigating how low-pressure systems may change in the future

Post by StefanR5R »

See the other recent CPDN related threads in this News section: They are turning to models with much shorter task durations, but with big memory footprint and huge result files. Currently this is Linux only. Longer term they want to offer OS independent vboxwrapper based versions of this. (I.e., with even larger memory footprint and all the VirtualBox and wrapper related headaches.)

https://www.cpdn.org/server_status.php's bottom table shows what's happening. Mind the "Users in last 24 hours" column which tells which of the rows are current ( = those with more than a handful users in the last 24h) or obsolete ( = those with less than a handful users in the last 24h; the large numbers of tasks in progress of the obsolete applications is caused by their enormous result reporting deadlines).

Put https://www.cpdn.org/apps.php side by side to see which computer platforms can run which of those applications. — OpenIFS apps are the current ones, and only available on Linux for now. BTW, all of their older applications are 32-bit applications; OpenIFS is 64-bit.

Currently you need:
  • Linux,
  • 7 GBytes RAM per each running task, plus a decent amount of RAM for anything else which is happening on the computer outside of BOINC,
  • an internet connection with a really fat upload link.
  • If you set up a Linux VM for this, also take care to give a very large amount of disk space to the VM. Depends on the active work batch, and on how many tasks you want to execute in parallel, want to have in progress, and will occasionally be unable to upload for some periods of time because their upload server can't really handle this stuff adequately.
When they have work available, a computer cannot easily run very many tasks of this in parallel:
a) due to the RAM size requirement,
b) due to the need to upload huge amounts of result data per task,
c) related to a), because these models are very RAM access heavy, taxing the memory controllers more than the CPU's arithmetic logic units. The more of these tasks run at once, the slower they get.

Added in 16 minutes 40 seconds:
PS:
That's heavy duty physics simulations for you. Quite a different cattle of fish than the usual number-theoretic DC applications out there.

CPDN want to take these kinds of models to higher grid resolutions, which implies even bigger RAM footprints, more computing time per time step, and more time steps required for the same simulation period. (Smaller spatial distances between grid nodes implies that time steps need to be smaller too, in order for the simulations to converge.) Therefore they will eventually switch to multithreaded application versions. The code which they use is already prepared for this, and is already in use in this fashion outside of CPDN.
Post Reply