PRECIS support forum

It is currently Tue Mar 19, 2024 2:23 am

All times are UTC




Post new topic Reply to topic  [ 1 post ] 
Author Message
PostPosted: Tue Apr 15, 2014 2:56 pm 
Offline

Joined: Thu Feb 06, 2014 12:29 pm
Posts: 118
The real, "wall clock" time that it takes to run PRECIS experiments depends on many factors.

First of all, the choice of resolution (either 50km or 25km) obviously influences the rate, because there are more grid boxes at 25km over a similar domain (i.e. each 50km gridbox contains four 25km*25km grid boxes). Furthermore, the time step of the regional model is reduced accordingly at 25km. Ultimately, our rule of thumb is that a 25km experiment takes about six times as long to run as a 50km experiment over the same area.

Next, the size of the domain. A regional domain covering Mexico will obviously run more quickly than one over the whole of North America. You should be careful not to make your domain too big or too small. The PRECIS team should "sign off" on your domain before you start to run experiments.

Next is the length of the model run. The maximum run length (for HadCM3Qx driving GCMs) is 150 years (1949-2099). Obviously, if you simulate less years (e.g. 2000-2100) the completion of the experiment will be sooner. You should aim to run experiments for at least 31 years (1 year "spinup" that is discarded followed by 30 years of experiment).

There are other influences (e.g. whether using aerosol prognostics if available), but these are the major ones.


As for the computer you run PRECIS on, the speed of PRECIS is influence mostly by the CPU. The more powerful a CPU, the better. This includes multi-core systems (i.e. that a quad-core CPU is better because PRECIS can use all CPUs in parallel).

After the CPU is the amount of RAM (memory) in the PC. You should aim to have 1GB of RAM per CPU core minimum. If there is too little RAM in the PC it will slow down PRECIS. Note that the opposite is not true - PRECIS won't run faster with twice as much RAM.

Finally there is the amount of storage space. This does not affect the model's run rate per se, except that the main reason we see PRECIS crash is because the PRECIS output data fills up the entire hard drive. This is less of an issue nowadays with 1TB+ hard drives, but it can still happen.

In short, buy a shared memory Linux based system (see my sticky post on "Can PRECIS run on a cluster?") with the fastest CPU you can get, with at least 1GB of RAM per core (2GB would be even better), and lots of storage space (4TB is good). :)

Bear in mind that PRECIS tends to use 100% of the CPU power over all cores. If you are running other processes on the same computer, PRECIS will be slower.

D.


Top
 Profile  
 
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 1 post ] 

All times are UTC


Who is online

Users browsing this forum: No registered users and 6 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
cron
Powered by phpBB® Forum Software © phpBB Group