[HN Gopher] Implicit In-order Forests: Zooming a billion trace e...
___________________________________________________________________
Implicit In-order Forests: Zooming a billion trace events at 60fps
Author : trishume
Score : 19 points
Date : 2021-04-30 22:13 UTC (47 minutes ago)
(HTM) web link (thume.ca)
(TXT) w3m dump (thume.ca)
| quotemstr wrote:
| I don't like it when programs rely on VM overcommit in such a way
| that they break on other systems. Even if you don't care about
| Windows (which is a strict accounting no-overcommit system), you
| should care about Linux, which can be configured to do strict
| accounting the way Windows does.
|
| If you want to use the big address-space carve-out trick, you
| can, but the right way to do it is to PROT_NONE the parts of the
| address space you aren't using and install a signal handler to
| commit bits of your carveout as they're used.
| JoshTriplett wrote:
| Processing a signal and mapping memory via that signal is less
| efficient than letting the kernel do it for you.
|
| You'll get better performance by using userfaultfd.
| londons_explore wrote:
| Isn't this the idea behind MIPMAPS in computer graphics?
|
| In the tracing world, I believe the opensource pulseview/sigrok
| does this. It makes the UI very responsive even with gigabytes of
| data. I just wish it also integrated data compression of some
| kind so I could fit more trace than I have RAM (it can't be all
| that hard to intelligently compress a super repetitive signal in
| a way which still allows this fast zooming and scrolling -
| replacing some tree nodes with LZ77 style backreferences ought to
| do the trick)
___________________________________________________________________
(page generated 2021-04-30 23:00 UTC)