Hacker News Reader

Comment by xxpor | Hacker News Reader

Can anyone explain why the windows update process is so freakin slow?

It's just sits there scanning for ages, and then installs are really slow as well. Downloads seem fine. Is it not just asking for updates since x?

View on HN

Yeah, honestly, I'll believe it when I see when it comes to faster updates on Windows. Whenever I have to update macOS even on a machine that's way, way behind, it's super simple and straightforward. On Windows, there's always hours of pain and installing endless updates over and over. Windows Update never seems to finish updating. Right after I finish installing everything, boom, it's found more updates. It's just absolutely horrifying and I wouldn't wish that on my worst enemy.

> Right after I finish installing everything, boom, it's found more updates.

I've always thought that's because the newest ones are dependent on some of the second-newest ones, so it couldn't install them before the dependencies were installed. And the reason it throws those at you immediately, "boom", is that it hasn't really "found more updates" after installing the first set; it knew about these ones too, from looking through the list it made for the first update, and just held them back for a second round.

Just my speculation, though, so could be totally fucking wrong.

On Windows an admin can selectivly apply individual updates, which isn't possible on macOS, I think. The core algorithm that calculates the dependencies is horribly slow. They are working around that with update rollups and cumulative updates since Windows 10. There are multiple layers of snapshotting going on during updates (VSS + Transactional NTFS).

Then for example "Update telemetry" scans all applications on the system and only gives you the option to upgrade to Windows 11 if it doesn't find anything incompatible. On Apple you just get the option to upgrade and stuff just stops working afterwards.

> On Apple you just get the option to upgrade and stuff just stops working afterwards.

Granted, they host an "incompatible apps list" which their installers are bundled with + download updates if they can. I have no idea what it's for and what it's doing, never found a thing on my systems.

Microsoft's done away with "individual updates". The current update for Windows 10 is "October 12, 2021 Update" and aside from a few exceptions such as CPU microcode updates, it bundles every previous update.

Which certainly is easier to coordinate as a company. Part of the reason Apple does entire OS patches as that it allows every aspect of the system to have a consistent view of each other part, so API (private or public) are all in sync between all the apps and components. It also means full QA passes are for a consistent set of versions as unexpected bugs can impact far away parts.

That process solves for the problem of only wanting some updates. The new way you get all or nothing. The situation that you bemoan is exceedingly rare by comparison to updating in general. The point is probably moot though in light of the changes to Windows update over the years. There are plenty of reasons to hate on Windows, I think update is one of the few things that works reasonably well. The contents of those updates is another story.

Because you're still using Windows 7

Last time I got frustrated by Microsoft update time, I created a PR for their update documentation while I was sitting there watching it that described how long it took. They merged it.

https://github.com/MicrosoftDocs/OfficeDocs-Exchange/pull/27...

That genuinely made my morning.

On what version of Windows?

Windows 7 and earlier had some issues where, as the size of the update catalog grew, the updater would start taking much longer to "scan for updates". This was eventually fixed in an update to Windows 7 -- but this still meant that you'd be waiting a very long time to install the first few rounds of updates to a fresh 7 install.

That being said, Windows 7 is EOL. If you're still using it, please stop.

Windows Server 2019, for a VM running on a server that had nothing else to do, PCIe 4 nvme SSD, over 10 minutes waiting for windows update to do its thing on reboot.

I don’t mind so much that it takes time checking, downloading and installing updates, because that’s not downtime. My beef is with what happens once you reboot.

I have a new PC with a NVMe drive and Windows 10. Everything is fast on it, but a ~300MB monthly Windows update can easily take 2-3 minutes, plus a reboot. Usually a software install of this size would take no more than 20 seconds.

This has been my experience with Windows updates for the last 10 years or so, on different PCs and different Windows versions. It's just slow (and it's definitely the installation phase, not the downloading).

Good old accidental exponential time complexity. Not caught in the initial release as there were not enough pending updates to notice the issue.

And but of course nobody's going to make the effort to internally test with super outdated installations (although a VM snapshot or two, or a new install, would've made it simple enough). Good point.

quadratic is the worst IME. will get through moderate, perhaps even large test suites, sit in prod quietly waiting for the data set to grow and only then blow up.

And no tests.

Windows cannot overwrite files which are in use. I guess this complicates a bit the issue. But they are working on GUI "improvements" which nobody asks so it will take a while until they improve the situation.

It can rename file that are in use though.

Last time I looked at the update performance in windows 10 the issue was that the installers for each of the patches replaced files sequentially, that sequential execution in turn got delayed by single-threaded antivirus.

It could be a lot faster if they parallelized the file operations or deferred the antivirus.

I believe that BigTech started releasing huge updates to normalise having an always-on internet connection during updates and normalised the slow update process for the opportunity it provides them to mine and analyse all data on your computer. Often an update is the only time when they can (and do) reset and toggle all the privacy settings back to their favour, and then do on device analysis (like collecting meta data of all your files or analysing your photos etc.). They can then send the data back to their servers to add to your profile they have on you. That's why I always make it a point to immediately switch off my router after downloading an OS update and starting an install.

(Both Windows and macOS spend a lot of time in indexing your disks and doing all sorts of on-device "AI" analysis to provide some specific feature. And both collect a lot of your data for this.)

i think this is not far fetched

what linux distro do you recommend?

After several years ... I have come to the conclusion distro wars are pointless.

These days, I just have an old XUbuntu installation that has been consistently upgraded since 16.04, and it just works. Some may consider Ubuntu-clones to be "beginner's distros", ultimately, though, I like how it gets out of your way, apt works well, and I can focus on being productive.

Doesn't mean you have to use all they do. I happily work emacs-based and in i3...

thanks for that, it’s been years since I tried xubuntu and “just works” is a big selling point.

I prefer FreeBSD and Debian.

When I noticed installing or upgrading software was slow, and checked resource monitor, in 60% of cases it was windows built-in antivirus consuming all CPU time.

Yes, I have noticed that too. But the windows update process itself also sits there for minutes using 60%+ CPU.

Even installing...just doesn't make sense why they're so slow.

I have a pretty respectable system. i9-9900K, 32 GB of RAM, and Windows 10 is installed on an NVMe that can read/write at up to 3 gigabytes/sec (Yes, gigaBYTES!).

A 200 MB update should take a fraction of a second. Even if it's actually applying a delta and not just simply overwriting files, I can't imagine it should take more than a couple seconds.

This is a massive problem for large windows hypervisors. We have to shutdown around 20 vms for about 40 minutes for the host to reboot after updates.

We are looking to move to a Linux host instead but that is a big up-front cost to get the same size machine, move the VMs across and remove the old host.

Live VM migration is a thing on Hyper-V and host maintenance is its primary use case (as well as load leveling)

It's gotten a lot faster recently. What's still slow, though, is macOS's updates.

Historically there were CPU intensive operations, I assume to trade compute time for bandwidth. I am not sure how much of that still exists though since updates have become all-or-nothing blobs.

https://dilbert.com/strip/2010-03-29

Exactly. I can update linux in a matter of seconds, the longest part is downloading the updates.

My thought exactly. I would take a 2GB monthly update with a simple swap of the assemblies that would take 2 seconds and a reboot over the current agonising update process.

Honestly, only kernel updates should require host restarts. Linux and BSD have had this for ages. Maybe have a flag that security updates use to flag any process using a vulnerable DLL as needing an eventual restart, and then have an administrative setting for restarting vulnerable processes after a certain period of time, at a convenient hour.

> Linux and BSD have had this for ages.

Fedora/GNOME prefers to install updates during a reboot. So although most updates can work this way, it's not always the case.

There are some minor cases where a reboot is better, or where an update requires a bit of work. E.g. Firefox updates, but also I sometimes have that video playback suddenly breaks (not sure why; kernel/mesa/something). Only a reboot seems to fix the video playback and it happens in various applications once it breaks (Firefox, mpv). I install the updates via dnf via cron/systemd (forgot what I did), so not with the suggested reboot and so on.

I do appreciate the Flatpak bits, those update easily.

How would this work? For example there was an update in libXYZ and the office suite, browser, audio player and some system services need to be restarted to use the new version. Obviously the system can't just quit those applications, after all they are in use and there might be unsaved data. So should the users get a popup for every app that needs to be restarted? And what about the system services? E.g. when the display session needs to be restarted, all open applications would be lost as well, or restarting the audio server might mess with an ongoing audio recording.

Wouldn't it be so much more reliable, faster and simpler to just install those updates on the next restart?

Presumably the user would get a prompt telling them they need to log out and log back in to pick up security updates, with a "details" button that shows them which applications actually need to be restarted if they don't log out/log back in.

It's certainly simpler for the developer to pick up the updates only upon rest, but it's sometimes very inconvenient for the user, and probably slower to pick up the actual updates. For instance, I run a vanity domain from a Linux device at home. It would be inconvenient to run Windows on that machine and have to restart it following monthly patches or following emergency patches.

Interestingly I noticed KDE Neon went the Windows way, and now requires reboots for all their updates.

Before this the experience could be quite fragile post-update, so I tended to always reboot right away anyway.

Note that Linux can do kernel updates with a reboot.

I think you meant to say _without_ a reboot, no?

Yes :)

Windows used to support rebootless kernel updates too, but for some reason they stopped doing it.

kexec is for all intents and purposes a reboot though, if that's what you're thinking of. Granted, you skip the EFI/POST part but you still lose all of your state.

Otherwise, are there any distros regularly using ksplice?

Do you have a spinning rust disk?

It's slow even on my PCIe 4 NVMe drive. I boot into Windows once a week or two and I often see it spending more than 5 minutes installing a single update package. This is even slower than apt and dpkg, props to them.

Your installation of Windows is that by any chance the installation that came with computer? Or is it a Windows that you have upgraded from a previous version (and perhaps that was upgraded from yet another previous version)?

I never have these slow downs that I read others have. I always do a clean install from a USB stick (including deleting recovery partitions). I do this like once a year or 1 1/2 years when something is being released that is either a new version or what is we used to call a service packs.

Windows sucks at upgrading. There is always something strange going on after upgrading. Things that not happens after a clean install.

If it is looong time since you have performed a clean install, I recommend you to consider doing that.

I installed last year when I got the computer and drive, and it's still pretty clean since I don't really use Windows these days. Or was clean until I upgraded to 11 last week.

But this has always been my experience. I remember reinstalling Windows 7 and waiting hours for Windows Update to finish. Even checking for updates could take 20 minutes.

yeah, endless hours spend waiting for Windows Update. It is not so bad anymore.

The only thing I can think about comparing your setup to mine is that I rarely turns off my computer. If you only use your very infrequent and turn it off when not used. I wonder if Windows is doing all kinds of maintenance jobs every time you start your computer since it is off most of the time, and then when you manually hit windows update, it gets a bit busy working out the correct state of your computer

Those jobs are on my computer distributed over days / weeks but wit you they run every time you turn on windows since it couldn't run them at their scheduled frequency since the computer was off.

I am just speculating, but maybe try to turn on windows once in a while and leave it over night, to test if this improves things.

Not to diminish package managers, I don't think they are having to be quite as clever as Windows update, at least historically. I could be wrong, but I think they are more like dependency tree managers that also download and execute new installers.

No, they aren't clever. They're practical. There's no such thing as execution here. Packages are archives containing the entire files not differences. The manager downloads it, unpacks it, and adds/removes/replaces the package's files in the system.

They can also execute post-install or post-update scripts. And indeed, Linux package managers aren't as fancy as Windows Update is, but that fanciness isn't something I miss.

My distro's package manager probably breaks my system if it loses power during an update. Even so, it's been one of the more reliable and pleasant to use programs I've encountered.

That doesn't seem right.

I also have a PCIe NVMe boot drive, typically i don't even notice updates until i go to shut my PC down in the evenings and windows tells me it'd like a few mins to do the updates.

Same difference to me, i was shutting it down and walking away anyway

I notice them because I intentionally open Windows Update and install the available ones.

Windows updates are released once a month.

No, but where I notice it is on my work computer that is admittedly loaded down with corporate security nonsense. My personal computer doesn't have nearly as much of an issue, but on the other hand it's an always on desktop so it's doing most of its windows update work at 3 in the morning.

Interestingly, I still use magnetic drives as a second drive on my laptops for storing data.Unfortunately most modern laptops have only one drive (NVMe), excellent for performance and less so for durability. And you better not keep it off for a couple of months.

Reply on news.ycombinator.com