Paul Hauner

@paulhauner

Ethereum 2.0 at .

Sydney, Australia.
Vrijeme pridruživanja: srpanj 2009.

Tweetovi

Blokirali ste korisnika/cu @paulhauner

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

  1. 31. sij
    Poništi
  2. 17. sij

    I'm really looking forward to reading the results of this. I can't wait to see what they find. Congrats to the research team for getting to this point!

    Poništi
  3. 15. sij

    If you're a Rust developer, check our Twitter for the job offering. If you're a UI/UX designer, check for our validator GUI RfP. If you're a contracting devops pro, DM me directly because I need help launching testnets. If you have questions, feel free to ask :)

    Prikaži ovu nit
    Poništi
  4. 15. sij

    I said it before, but it's been crazy at . We're absolutely giving eth2 all we have and it's going to happen.

    Prikaži ovu nit
    Poništi
  5. 15. sij

    Additionally, we've been feeding back to the specification and the EF researchers have made some major changes to the eth1 voting mechanism based on our feedback about optimizations.

    Prikaži ovu nit
    Poništi
  6. 15. sij

    There's also been tons of work happening in the networking stack and the rest of the client; all the core team have been doing amazing work. On top of the actual dev work we're also interviewing Rust developers, preparing RFCs for GUIs and scheduling our security review.

    Prikaži ovu nit
    Poništi
  7. 15. sij

    We already know how we can get rid of that memcopy, we just need to get around to doing it. There's honestly so much going on that there's low-hanging fruit we haven't yet picked. I think we can drop those state hashing times too, but that's no so low-hanging.

    Prikaži ovu nit
    Poništi
  8. 15. sij

    I think Herumi will be very helpful because roughly 50% of our block processing time is signature verification. Another ~30% is merkle-hashing the beacon state and other ~15% is just doing memcopy of the beacon state and all it's caches.

    Prikaži ovu nit
    Poništi
  9. 15. sij

    The Herumi library looks like its ~2x faster than our native Rust library, and we're not even using it yet. We're saving that one for later.

    Prikaži ovu nit
    Poništi
  10. 15. sij

    . has been working on some BLS optimizations. We asked him to spend a day on seeing if he could make our Rust BLS library faster and he managed an easy 10%. Additionally, he's integrated the super-fast, hand-written assembly BLS library by Herumi.

    Prikaži ovu nit
    Poništi
  11. 15. sij

    As a result of these optimizations, we're seeing Lighthouse syncing a 16k validator, production spec network at somewhere between 6-20 blocks/second. 6 b/s is what our t2.medium AWS nodes do and 20 b/s is what my flashy desktop does. We're still improving too!

    Prikaži ovu nit
    Poništi
  12. 15. sij

    That trade-off is configurable at the CLI, so users can choose if they'd just like to sync fast and don't care about fast access to historical info (e.g., validators), or if they don't care about syncing but want fast access to history (e.g., block explorers).

    Prikaži ovu nit
    Poništi
  13. 15. sij

    In fact, we never actually lose historical information. Instead, it takes longer for us to retrieve that information. So, we have a trade-off between syncing fast and retrieving historical information fast.

    Prikaži ovu nit
    Poništi
  14. 15. sij

    We've also had massive gains on syncing due to heavy database optimization by . We're talking 4x improvements, at least. Michael has massively reduced the amount of bytes we need to write to disk whilst syncing, without ever forgetting any history in the chain.

    Prikaži ovu nit
    Poništi
  15. 15. sij

    The new fork choice algorithm is up to 40x faster in some areas! Adding a new block to it takes literally microseconds now. Finding the head takes less than a millisecond on my laptop. Additionally, there's 1/2 the amount of lines and 10x more testing. Improvement on all round!

    Prikaži ovu nit
    Poništi
  16. 15. sij

    Luckily, we've been hardening against loss of finality lately so the testnet pulled through without loss of data. To fix the fork choice problems, this week I've fully re-written our fork choice code based upon some great research by .

    Prikaži ovu nit
    Poništi
  17. 15. sij

    Last week we released a v0.1.1 release, but a bit hastily. One of the database optimizations didn't play well with our fork choice algorithm and we saw testnet instability. A bunch of validators started producing on weird forks and we lost finality several times.

    Prikaži ovu nit
    Poništi
  18. 15. sij

    Firstly, we've been running a public testnet for almost 4 weeks now without really announcing it. We've been so busy re-working and optimizing that we've never actually felt comfortable releasing it. The next big improvement is only one more PR away, so we keep delaying it.

    Prikaži ovu nit
    Poništi
  19. 15. sij

    It's been crazy at Lighthouse over the past month, I've barely had time to check Twitter, let alone post! We've been focused on optimizing sync times and have seen lots of improvement. Here's a little recap on what I can remember from recent history 👇

    Prikaži ovu nit
    Poništi
  20. 15. sij

    We're looking to engage a team to build a front-end for the Lighthouse validator client :)

    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:

    ·