Thomas Munro

@MengTangmu

PostgreSQL and FreeBSD hacker, 🇪🇺🇬🇧🇳🇿

Wellington, New Zealand
Vrijeme pridruživanja: ožujak 2013.

Tweetovi

Blokirali ste korisnika/cu @MengTangmu

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

  1. 25. sij

    I wonder how hard it would be to jam an ARM board and a RISCV board into an HP Microserver G8 case, for build/test farm work, for tidy power/network/cabling in a cube form factor that is stackable with the rest of my homelab...

    Poništi
  2. 21. sij

    I have a btree scan prefetch patch which works pretty well in some cases like this, but I didn't think of the vismap pages. Could add that but now I am wondering if we shouldn't just prefetch the *whole* vismap...

    Poništi
  3. 20. sij

    Drat. I had a proposal lined up but it seems the CFP closed a bit earlier than I expected (I expected the *end* of the 19th in Ottawa). That'll teach me to leave things to the last minute.

    Poništi
  4. 17. sij

    Day ... erm... 7/7.

    Prikaži ovu nit
    Poništi
  5. proslijedio/la je Tweet
    14. sij

    We propose Pivoted Query Synthesis (PQS), a new approach for finding logic bugs in DBMS (see ). Using PQS, we found ~100 previously unknown (and many critical) bugs in widely-used DBMS (e.g., SQLite3, MySQL, and PostgreSQL). Work with .

    Prikaži ovu nit
    Poništi
  6. 14. sij
    Prikaži ovu nit
    Poništi
  7. proslijedio/la je Tweet
    13. sij
    Poništi
  8. 12. sij
    Prikaži ovu nit
    Poništi
  9. 12. sij

    Day 4/7. I challenge . 7 days' photos from your life, no people, no comments.

    Prikaži ovu nit
    Poništi
  10. 12. sij
    Prikaži ovu nit
    Poništi
  11. 9. sij
    Prikaži ovu nit
    Poništi
  12. 8. sij

    Day 1/7 No challenges, if you feel challenged, act accordingly!

    Prikaži ovu nit
    Poništi
  13. 8. sij

    Does "make" have the right kind of information to be able to speed up large build jobs with POSIX_FADV_WILLNEED?

    Poništi
  14. 25. pro 2019.

    How to achieve context-switch-free async I/O à la , on ? Could kqueue support an optional user space output queue? For the request side, I wonder if a new aio multiple-request syscall called from a user thread might be about as good as what io_uring does.

    Poništi
  15. 22. pro 2019.

    Here it is for ZoF (instead of the legacy ZFS in the tree today). Not sure if this is sane enough to propose beyond Twitter yet (is the locking OK?) but it'll let me do my current PostgreSQL work on my favourite OS so there's that!

    Prikaži ovu nit
    Poništi
  16. 21. pro 2019.

    Ok another late night hacking session, and the answer so far appears to be: yes! About a 4x speed-up for random cold cache pread() stream on my laptop, by using posix_fadvise() to prefetch 20 blocks ahead.

    Prikaži ovu nit
    Poništi
  17. 18. pro 2019.

    Dependency on chocolatey removed for now. That should clear up this new wave of bogus build failures.

    Prikaži ovu nit
    Poništi
  18. 14. pro 2019.

    I spent an evening poking around in the ZFS source, and connecting dmu_prefetch() up to POSIX_FADV_WILLNEED. So far I can tell you that it compiles, and doesn't crash my kernel (woo!). Tune in next time when I figure out if it's really prefetching :-)

    Prikaži ovu nit
    Poništi
  19. 8. pro 2019.

    I should add that I don't know what I'm doing and there may be basic things wrong with that code but it *seems* to work in quick tests. Testing with a set of not-yet-posted PostgreSQL patches to sprinkle PrefetchBuffer() in more fun places to reduce stalls.

    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:

    ·