Login
You're viewing the assemblag.es public feed.
  • Sep 10, 2026, 5:16 PM
    pssst

    hey

    parallelizing your init system is stupid
    💬 1🔄 0⭐ 0

Replies

  • 💬 0🔄 0⭐ 0
  • Sep 10, 2026, 6:53 PM
    How much time would you say your computer spends 'booting' as opposed to 'booted'? How much time does parallelizing that minute window of operation save you?

    I have server nodes that take 20 minutes just to POST. The actual init system is meaninglessly small compared to that.

    The only class of people who might really benefit from fast boot times are init system developers who have to boot all the time for testing.
    💬 1🔄 0⭐ 0
  • Sep 10, 2026, 7:54 PM

    @khm @rl_dane

    The only exception I can think of, maybe, is laptops that POST quickly (a few seconds tops) but don't support suspend (like, say, plan9), and which are frequently travelling and often not near outlets (which is why when I was travelling frequently I was trying to get my thinkpad to boot in like 20s instead of 100s [it was mostly slow because of me, not 9front upstream!], but which is also why, now that I'm not, I kinda just go... yeah this is fine)

    Actually, one more current example for me: I was taking my laptop in to work for when I was on break, spending 30 minutes on pdffs

    If it takes 3 minutes to boot, that's 10% of my time gone. If it takes 30 seconds, that - sounds like nothing, but when the time window is so small already, and I often hit the "oh I get it now!" state 20-25 minutes in, those three minutes are the difference between, the next time, sitting down and going "ah fuck, this is half broken and I don't remember what I was doing" vs "oh cool, it's already committed!"

    But, I will point out that all I really needed to do in order to boot faster was switch the file system :P

    💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • Sep 10, 2026, 8:02 PM

    @khm @rl_dane

    Well yeah sure.

    Though I suppose technically if I patched init to run oldventi immediately on startup... would only be useful on my machine of course but would probably shave off like 10 seconds tbh.

    (I still use venti for some data, even as gefs is $home on the laptop now.)

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Sep 11, 2026, 2:44 AM
    3TB system RAM and 2.3TB GPU RAM across eight GPUs. I only have a few hundred of these tho
    💬 0🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • 💬 1🔄 0⭐ 0
  • Sep 11, 2026, 3:37 AM

    @khm @rl_dane

    were you also the one who previously worked with xeon phi or was that kvik? :P

    you both have short k- names and I haven't interacted with either of you in person so my brain put you both in "online 9 k people" I think :)

    💬 2🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Sep 11, 2026, 12:08 PM
    yep that's me as well! I don't know why people pronounce "NIC" as "nick" but "MIC" as "mike." Fortunately the entire xeon phi product line evaporated so the disparity is resolved.

    CC: @rl_dane@polymaths.social
    💬 1🔄 0⭐ 0
  • Sep 12, 2026, 5:13 AM

    @khm @rl_dane

    On the other hand, as far as memory bandwidth per dollar goes, AFAICT the ones on ebay are quite possibly the best option on the market, period. Better than any GPU or computer.

    💬 1🔄 0⭐ 0
  • Sep 12, 2026, 5:21 AM

    @khm @rl_dane

    [...I'm kinda shocked that Certain People haven't noticed this and started using them for Certain Things.

    I've... kinda been dabbling at getting a 9p-over-pci interface working for it so I can do some custom ML stuff, tbh.]

    💬 1🔄 0⭐ 0
  • Sep 12, 2026, 5:22 AM

    @khm @rl_dane

    > 9p-over-pci interface

    which, btw: fairly sure hardware can do x16, not just x1. I think it might _require_ rebar/4g decoding, though, same as their consumer GPUs.

    It's also possible that MPSS is just trash, but I suspect bad / incompatible firmware might've been a factor too

    💬 1🔄 0⭐ 0
  • Sep 12, 2026, 6:09 AM

    @khm @rl_dane

    also also also I would really, really, really like them to have done a good job on the xeon phi

    I can simultaneously accept that the xeon phi objectively is shit, that I'm silly for liking it, while also arguing that a _good_ implementation of it would have been _amazing_ and I _still want that_

    💬 1🔄 0⭐ 0
  • Sep 12, 2026, 4:51 PM
    I'm on board with this. It was a good dream. If you can get your hands on a Knight's Landing machine it would be worth playing with. Unfortunately "having a good idea, rushing it to market, and the rugpulling it midway through the product cycle" is what Intel specializes in

    CC: @rl_dane@polymaths.social
    💬 1🔄 0⭐ 0
  • Sep 12, 2026, 11:39 PM

    @khm @rl_dane

    There's a few people selling the CPUs for those at reasonable prices - I could probably get 3 of them for under $100 total! - but the boards...

    💬 1🔄 0⭐ 0
  • Sep 13, 2026, 4:48 AM

    @khm @rl_dane

    > "having a good idea, rushing it to market, and the rugpulling it midway through the product cycle" is what Intel specializes in

    I also think it's kinda funny that, at their peak when they had a buncha money, they probably lost more by fucking up like this and trying to cut their losses than they would have by just throwing money at stupid ideas

    they would have better off putting out more stupid products, figuring out how to make them good, and supporting them...

    And then they fucked it up so badly that now they can't even afford to do that

    💬 0🔄 0⭐ 0