Tweets

You blocked @danluu

Are you sure you want to view these Tweets? Viewing Tweets won't unblock @danluu

  1. Pinned Tweet
    24 Dec 2017
    Show this thread
    Undo
  2. Jan 22

    I'd like to link to the tweet in a blog post that describes the version of the system you normally only hear about at the bar as a justification for discussing the "real" system, warts and all, but I haven't figured out the exact search I need to return the tweet :-(.

    Show this thread
    Undo
  3. Jan 22

    Can you help me find a tweet? Someone tweeted something like "talks describe an idealized version of the system, if you want to know what the system was really like, you have to talk to the speaker after the talk and public Q&A". Who tweeted that and/or what's the link?

    Show this thread
    Undo
  4. Jan 5

    Algorithms interviews: theory vs. practice

    Undo
  5. 30 Dec 2019
    Replying to

    Google Design's "Best of 2019" page is 64MB, takes 98s to load the "2019" that's the main visual element above the fold (from a Cable modem in Australia).

    Waterfall graph of page loading. The giant above the fold gif is the first thing to start loading, but it's overwhelmed by the other huge gifs that load in parallel.
    Timeline of page updates, it's 98 seconds before the above the fold gif goes from showing the top half of an outline of "2019" to all of an outline "2019".
    Undo
  6. Retweeted
    11 Jun 2019
    Undo
  7. Retweeted
    17 Jun 2018

    Restartable sequences in Linux: . This is useful for implementing fast, per-cpu data in user-space. When there are large numbers of threads, some uses benefit a fair amount from per-cpu data over thread-local data.

    Undo
  8. 15 Dec 2019

    Interesting indirect argument for pair programming

    Undo
  9. 12 Dec 2019

    Perhaps not coincidentally, at the place where the vast majority of candidates weren't rejected for criteria that are weakly or uncorrelated with job performance, we didn't think about hiring much because it was so easy That company had the highest average productivity I've seen

    Show this thread
    Undo
  10. 12 Dec 2019

    Funny, managers at all but one company I've worked for say this while, simultaneously, most referrals are dropped on the floor before the technical interview and the vast majority of reasonably qualified candidates are rejected in interviews.

    Show this thread
    Undo
  11. 10 Dec 2019

    To be clear, unlike that commenter, I appreciate packaging work since it often lets me run binaries with little work on my part (in another thread, quoted OP says distro packagers are useless), but I thought it was interesting to see why using Jekyll was so high overhead for me.

    Show this thread
    Undo
  12. 10 Dec 2019

    I would regularly try to publish a post or even just fix a typo only to find that I had to install a bunch of packages, override environment variables, etc., to get around what were effectively compiler or linker errors (not that people call them that in a dynamic language).

    Show this thread
    Undo
  13. 10 Dec 2019

    Huh. I guess this is why it was so much work to maintain a working Octopress/Jekyll install back when I used Octopress (speaking as a non-Ruby dev who didn't maintain a Ruby dev environment and relied on system packages as much as possible)?

    Twitter's alt text limit is too short to quote the first entire sub-bullet of this with any context, see https://lobste.rs/s/6ame3m/developers_shouldn_t_distribute_their#c_ozfhi8 for the comment I've quoted and https://lobste.rs/s/6ame3m/developers_shouldn_t_distribute_their for the entire set of threads.
    Show this thread
    Undo
  14. 8 Dec 2019

    BTW, if you want to try CPUID spoofing without virtualization and have a VIA processor, Agner Fog wrote this little utility: Performance delta on benchmarks varies, here's an example of a ~50% gain (47%):

    Score with normal CPUID: 1845
Scored with spoofed CPUID: 2721

(higher is better)
    Show this thread
    Undo
  15. 7 Dec 2019

    The Google Chromium team banning our CPUs is especially ironic in retrospect since they cited security concerns. At the time, we were mostly shipping in-order CPUs, not vulnerable to Metldown/Spectre/etc. and of course Intel is the most vulnerable these.

    Show this thread
    Undo
  16. 7 Dec 2019

    That post blames Intel, but it's not just them. NaCl forcibly crashed (and Google refused our trivial fix!!!), random drivers wouldn't work, you had to get a patched Windows installer for multiple releases of Windows (difficult in the days of CD installers), etc.

    Show this thread
    Undo
  17. 7 Dec 2019

    This kind of thing is why the CPU startup I worked for allowed users to change the CPUID: you get huge performance gains from putting GenuineIntel in CPUID(0), but it would be a copyright violation to distribute our CPU with GenuineIntel in the CPUID

    ... The short story is that Intel checks for "Genuine Intel" CPU's when it's numerical library MKL starts executing code. If it find an Intel CPU then it will follow an optimal code path for maximum performance on hardware. If it finds and AMD processor is takes a code path that only optimizes to the old (ancient) SSE2 instruction level ...
    Show this thread
    Undo
  18. 6 Dec 2019
    Replying to

    A system for coordinating U.S. nuclear weapons moved off of 8-inch floppies earlier this year, 2 years late. Coincidentally, the original deadline was around when Google Cloud's product roadmap PM said that 1 year deprecation should be enough for anyone.

    Undo
  19. 2 Dec 2019

    If you're looking for a backend/infra role that: • has a high degree of autonomy • is collaborative (both intra-team and inter-team) • cares a lot about code quality • is serious about mentorship then this is one of the best run teams I know of, maybe the best.

    Undo
  20. Retweeted
    30 Mar 2018

    HOT new programming interview question: explain why Linux computes the maximum of two longs this way

    Show this thread
    Undo
  21. 20 Nov 2019

    How secure is hardware disk encryption? "In order to to recover the data from a locked MX100 drive, we connect a JTAG debugging device. Then, we use it to modify the password validation routine in RAM so that it always validates successfully"

    Show this thread
    Undo

Loading seems to be taking a while.

Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.

    You may also like

    ·