1. 35
  1. 5

    The author advocates using E-Prime (not a bad idea) when discussing performance comparisons, but the example used…

    on hardware H with T threads running in tight loops performing operations O, our specific lock-free queue has been measured to achieve a latency distribution of X1 and our specicific mutex-backed queue has been measured to achieve a latency distribution of X2…

    … violates the rules of E-Prime. Instead

    … may achieve…

    could work where

    … has been measured…


    Despite this one, very small nitpick, the article managed to accurately and -dare I say expertly- describe solutions to a range of system optimization problems that I wish the rest of my team could even begin to acknowledge.

    1. 5

      Indeed, e-prime does not come naturally to me (or anyone I’ve talked to about it) but I’ve found that wonderful benefits can be gained by including more context. The benefits very much follow a power law, in that by trading equivalences for context yields a very high initial return on investment, but the marginal value drops off, and speaking purely in e-prime may even reduce your ability to communicate effectively if applied fully, compared to partially.

    2. 8

      This is fantastic.

      1. 3

        Nice to see you sharing all the knowledge that you got from developing sled

        1. 4

          If you get past the intro, with it’s strangely aggressive and negative tone [0], the writing style calms down. I didn’t read the whole article (it’s long) but it looks like there’s some good info in there about how to measure (and present measurements of) performance. It does seem to jump between higher-level ideas (don’t just use averages; flamegraphs can be helpful) and lower-level ones, I think perhaps there could be some better organisation there.

          If the author is reading this I’d encourage them to reword (or just drop entirely) that intro. Maybe it’s supposed to be funny, but I found it overlong and off-putting, and I think it does a disservice to the rest of the content.

          [0] Some quotes from the intro, for reference:

          seriously, it’s always your fault if your engineers quit.

          You couldn’t give two fried pennies about computers if they didn’t help you feel the things you so desperately need to feel.

          Your entire life is entombed in an elaborate, entirely abstract labyrinth of machines that define so many aspects of every moment of your waking life.

          This guide brings together ideas [..] with the goals of helping you to be less of an asshole

          1. 4


          2. 2

            As part of the stated audience— I’m a Rust fiddler, new to performance-critical software— this is really great, made me laugh.

            1. 1

              FYI - the link for “The Tail at Scale” is broken

              1. 2

                thanks! fixed

              2. -4

                Wtf is this?

                1. 9

                  Website (internet)

                  1. 2

                    I don’t know why, but this was funny enough to make me spit my food out.