Hacker News Reader

Comment by jaymoore | Hacker News Reader

Speaking of Windows Update, does anyone know why it's so terrible?

On your average Linux-distro:

  Run "package-manager update"
  wait for download (ususally pretty quick)
  installs
  might need to restart if there's a kernel upgrade or something
  done.

But on Windows I dread every time there's an update because it's such a shitshow.

Download is slow - hobbyist running their own servers have better bandwidth than a trillion dollar company.

The install is slow.

The install doesn't even finish in the control panel dialog.

There's a second install step just before you restart.

There's a THIRD install step just before you login.

You go to Windows Update again to double check and guess what, it didn't install everything for some reason, repeat the above steps N fucking times.

View on HN

The install being slow is maddening. I used to have a stupidly overpowered Windows workstation I kept offline during analyses. When I would go to patch it, Windows patcher would only daintily sip available resources, and would take incredibly long to complete. I assume this was some directive to keep people from complaining about how much juice it takes to patch in the background, but if I manually click the button that says, “Patch Now”, that should mean engage Ludicrous Speed.

I read this blog post[1] a while back about the Windows memory subsystem performing pathologically bad due to an application (Google Chrome) asking the OS to employ memory throttling, and then blowing past its own self-imposed limits. I had a bit of a chuckle because I've literally run into, and complained about this before, so getting to see how silly it was under-the-hood was great schadenfreude.

These "background-friendly" processes are a sick joke IMO. My experience is they are wasting more power trying to play these stupid games, because by increasing the wall-time it takes to finish their workload, all they really achieve is keeping the CPU in a frequency-boosted state longer than necessary.

Just go fast and let the scheduler throttle you. If the browser updater is making a user-interactive system fall over: their code needs to be better, and the OS kernel needs to be better. If you are doing too much work the answer is "do less work" not "load-smear the frankly ludicrous amount of CPU and I/O." I think it's very indicative of how far desktop software has fallen that people seem to think adding more complexity (in the form of user-mode scheduling) is going to dig them out of the performance hole they dug themselves into.

[1]: https://randomascii.wordpress.com/2023/10/01/32-mib-working-...

IIUC windows can't replace currently running binaries, so windows first downloads everything needed for the update, then on reboot it will copy everything to the correct place and then reboot. While on linux your package manager downloads the packages, extracts them and thats it. On some distros that don't support live patching you'll need to reboot after a kernel upgrade.

it is quite amazing that one technical decision on locking is responsible for most of windows' unfixable problems

Are the Linux package manager binary updates (meaning the file updates themselves) transactional somehow? What happens to running processes? Presumably they need to be restarted.

Just overwriting files does not seem sufficient, given my limited understanding…

On linux a file that's open won't get deleted. It'll stay in memory. When you replace a file all it does is point the filename to the new file contents.

Any program open will keep running. Any newly started program will run the new binaries.

Of course this can give trouble with complex programs. Old executables seeing new config files, that sort of thing. But in general this doesn't seem to be a massive issue, and is fairly doable to work around by keeping critical files open all the time.

Thanks! I’m surprised that doesn’t cause more issues (especially with unattended updates on servers), e.g. programs not getting restarted when an update to a library is made, or newly started programs loading incompatible (some combination of old and new) dependencies.

I mean any program code is gonna stay loaded, so that's not an issue. So you're basically only left with resource files potentially causing issues. Most programs won't re-load their configs without getting prompted so that's not a problem either.

I feel like it's one of those things that just happens to cause surprisingly few issues and where they do happen it's either no biggie or gets fixed.

Reply on news.ycombinator.com