Jonathan Dursi

@ljdursi

I help groups in the broader public sector exceed at R&D with computing and data. Federated systems, teams, and organizations. jonathan@dursi.ca

Vrijeme pridruživanja: lipanj 2008.

Tweetovi

Blokirali ste korisnika/cu @ljdursi

Jeste li sigurni da želite vidjeti te tweetove? Time nećete deblokirati korisnika/cu @ljdursi

  1. Prikvačeni tweet
    29. ožu 2019.

    Grad students! Postdocs! New PIs! I need you to practice some lines. “No, I can’t do that.” “No, we don’t do that here.” (Thread)

    Prikaži ovu nit
    Poništi
  2. 28. sij

    Most of us know all this already; but non-simulation people can get tripped up when trying to incorporate (or review!) simulation work - and, increasingly, data science model-building work. This is a nice reference to keep in one's back pocket for those occasions 5/5

    Prikaži ovu nit
    Poništi
  3. 28. sij

    We can run exploratory simulations to inductively build a higher-level model of part of the system being simulated (e.g. how galaxies react under near-collision simulations) and even test hypothesis of that higher level model (under the model of the simulation software). 4/5

    Prikaži ovu nit
    Poništi
  4. 28. sij

    With simulation software we can do _both_ model building _and_ hypothesis testing (under the model), as long as we're clear which we're doing. 3/5

    Prikaži ovu nit
    Poništi
  5. 28. sij

    Simulation software is a model, not a hypothesis; it is built on, its implementation is verified and its runs are validated. We find limits of its range of validity, we don't disprove it. 2/5

    Prikaži ovu nit
    Poništi
  6. 28. sij

    The Cell paper cited below on hypothesis vs models is worth reading in the context of research computing and simulation broadly to help keep clear in our mind what simulation software is and what you can do with it... 1/5

    Prikaži ovu nit
    Poništi
  7. 24. sij

    (ok, 4/3): We actually know a lot about how to design governance of nonprofits, even federated nonprofits. The fact that not only are we choosing to not learn from that expertise but to also not learn from our own mistakes is just infuriating.

    Prikaži ovu nit
    Poništi
  8. 24. sij

    Setting up governance without knowing what the role and responsibility of the organization is, is a mistake anyway; ensuring the key accountability is to orgs with a clear conflict of interest (supporting their researchers vs winning contracts and capital funds) is absurd. 3/3

    Prikaži ovu nit
    Poništi
  9. 24. sij

    Yes, Compute Canada v3.0 was particularly broken, but it got that way and remained that way because the membership refused to get involved until the very end. 2/3

    Prikaži ovu nit
    Poništi
  10. 24. sij

    NDRIO (Compute Canada v4.0) will fail for the same reason versions 1-3 failed; the members should be the researchers who need the ecosystem, not institutions angling for capital funding. 1/3

    Prikaži ovu nit
    Poništi
  11. 22. sij

    Don't treat the cloud as a University machine room: imposing academic cluster structures on much more flexible capabilities while spurning 100s of productive managed services is not the way to efficiently deliver services *or* effectively serve specialized needs 7/7

    Prikaži ovu nit
    Poništi
  12. 22. sij

    The option that *doesn't* make any sense is to take the rigid structures of the "let's offer as many uniform cycles & bytes for as little cost as possible" approach and impose it on the expensive-unit-cost but super flexible and elastic commercial clouds. 6/7

    Prikaži ovu nit
    Poništi
  13. 22. sij

    But there's real justifications either way. The big uniform cluster is easy to reason about and monitor, and is cheaper for _some_ use cases. The lots of specialized solutions provides better service for the researchers but requires a lot of staff time to set up. 5/7

    Prikaži ovu nit
    Poništi
  14. 22. sij

    Both require a lot of effort expended - one to make custom solutions , the other to fit a round problem into a square queuing system. Specialized by research computing staff vs one-size-fits-most with researchers and trainees doing the work. 4/7

    Prikaži ovu nit
    Poništi
  15. 22. sij

    Academic clusters are about $. You buy one (or a handful) of compute & storage configs and use one (or a handful) queuing policies to try to compromise across the needs of high-need use cases as best you can. Researchers do what they need to do to make it work if they can. 3/7

    Prikaži ovu nit
    Poništi
  16. 22. sij

    The big advantage of commercial cloud for research computing - besides that there's just a lot of it - is the enormous flexibility in computing models, computing hardware, and research computing staff can put it together to serve the needs of the project. 2/7

    Prikaži ovu nit
    Poništi
  17. 22. sij

    I didn't comment on this blog post from a couple of weeks ago because I thought its main points went without saying in our research computing community but I'm sorry to say I was wrong... 1/7

    Prikaži ovu nit
    Poništi
  18. proslijedio/la je Tweet
    19. sij

    I’ve been asked what’s the primary difference between programming in astrophysics vs. programming in robotics. Here’s a thread on the differences I’ve observed, even though I programmed in the same language (Python) for both:

    Prikaži ovu nit
    Poništi
  19. 20. sij

    Upon reflection, I might have unfairly dismissed concerns about our Christmas tree staying up too long this year.

    Poništi
  20. 6. sij
    Prikaži ovu nit
    Poništi
  21. 6. sij

    Here's a page describing the initial plan for the newsletter, and a first signup.

    Prikaži ovu nit
    Poništi

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·