Alexey-Ben meeting 2019-02-19

[ashiklom]

Updates from last week

• FoRTE
  • ED ensemble outputs
  • Link to parameter values -- done
  • PFT-specific outputs -- done
  • Code for reading ensemble output -- evolving as I go, but converging on a workable set of approaches/functions
  • ED benchmarks -- done
  • When 4 ensemble members are running together, 1 simulation year takes ~30 minutes (so 100 years = ~50 hours)
  • With 6 ensemble members, slows down to 1 simyear per 45 minutes (100 years = ~79 hours)
  • After looking at top output, there is definitely a load-balancing issue. With 6 ensemble members, the system load average is ~24 (so the 8-core system is overburdened by 3x), and the ED processes alternate frequently between running and sleeping.
  • I asked IT if they had any suggestions. Here is the JIRA ticket link.
  • Thanks to IT recommendations, figured out how to run PEcAn across multiple virtual machines, so should be able to use a lot more cores (32 at a time) for ED runs (easily scalable to more). Spent part of Friday, and most of Monday and Tuesday on this, but now have a solid, scalable system where we can quickly add/remove compute nodes at will.
• Hector -- Didn't work on.
• Other
  • PNNL administrative tasks (outside activity form, CC application)
  • GCB re-review (not one of the other two I mentioned) -- ~30 minutes (all my comments were well taken)

Goals for this week

• FoRTE
  • R tutorial for FoRTE meeting
  • Finish PEcAn multi-node setup
  • Validate ED parameters
  • Start ED ensembles with finite canopy radius model
  • Clean up code for processing ED ensemble output (use drake more)
• Hector (time-permitting)
  • Literature values for permafrost "biome" pools and parameters

Other discussion topics

• At some point (probably after FoRTE Richmond meeting), I will need to take ~2 hours to resubmit trait meta-analysis paper somewhere.

[bpbond]

Thanks to IT recommendations,

Great! They're very responsive and helpful in my experience.

[ashiklom]

Update: Partway through FoRTE drake/git lesson. Here is the repo.

[bpbond]

How long are you thinking this will be? There's already quite a bit of material there.

[ashiklom]

How long were we planning for these lessons? I can adapt the lesson to fit whatever time frame we want. From earlier conversations, I thought our two lessons combined would be roughly half a day (~4 hours), but I can cut back if it's actually a lot less than that.

Also, even though there's a lot of text in there, I think we can move through the underlying steps fairly quickly.