I am running 120 hour back-trajectories using (concatenated) NARR meteorological data on the GUI. Using the daily run tool, I hope to run 10 years of back-trajectories at one location at 6-hour intervals.
However, each back-trajectory calculation takes a significant amount of time, generally 2-3 minutes, making this total run time-prohibitive (with the total run taking an estimated 15-20 days).
Is this run-time typical for this sort of HYSPLIT back-trajectory calculation? And if so, I assume it’s just a matter of finding a machine with more memory? I am using a 2017 MacBook Air with 8 GB of RAM, if that’s relevant.
It does seem like the run-time is pretty unpredictable, with back-trajectory calculations sometimes taking only a few seconds each, or other times taking >5 minutes. Often the program will freeze and give the Mac spinning “beach-ball” for a few minutes, delaying the results. Leaving HYSPLIT running over-night doesn’t seem to improve the calculation speed.
Any help would be much appreciated.
Back-trajectory calculations taking significant time
-
Harris_MRP
- Posts: 2
- Joined: September 3rd, 2022, 12:43 am
- Registered HYSPLIT User: No
Re: Back-trajectory calculations taking significant time
If you are having trouble with computation time, and you're using a standard hysplit install (i.e., not parallel), the most likely culprit is your processor's single-core performance. Hysplit will only use a single CPU core, and there isn't much that can be done to speed it up for a set hardware configuration beyond optimising the inputs to the model.
However, you can run at least as many instances of the hysplit executables (e.g., hyts_std) as you have CPU cores (see viewtopic.php?t=1260). You will need to make a working directory for each instance, and each instance will require its own chunk of memory for met data. I have not tried to do this with the GUI, but it works well via scripting. I recommend running as many instances as you have cores, minus one to leave some processing power spare for system processes. Running 4 or 5 can require quite a lot of memory.
However, you can run at least as many instances of the hysplit executables (e.g., hyts_std) as you have CPU cores (see viewtopic.php?t=1260). You will need to make a working directory for each instance, and each instance will require its own chunk of memory for met data. I have not tried to do this with the GUI, but it works well via scripting. I recommend running as many instances as you have cores, minus one to leave some processing power spare for system processes. Running 4 or 5 can require quite a lot of memory.