Hacker News Reader

Comment by GordonS | Hacker News Reader

> Windows 365 provides an instant-on boot experience

Unless they mean it'll instantly show a loading screen, I'll bet it doesn't.

I've used various VDI implementations over the years, and they've all been pretty slow to get to a working desktop.

I also wonder about latency. I have a 100/20 FTTC connection at home, which I find to be fast. Most recent VDI I used was Citrix, where I was expected to code in a VDI hosted in Netherlands (I'm in the UK) - the latency when typing drove me fucking NUTS! It was mostly just long enough to be intensely irritating and distracting, but occasionally latency would increase to a few seconds too.

View on HN

I decided to reject Citrix or Remote Desktop dev contracts on matters of principle, because you never know what irrational workflow they're coming up with next. If your customer/employer values their control-freakery higher than dev efficiency or sanity, I'm out. Recently, I had a contract where they used Citrix to work on a stupid Windows VM, only to execute Docker containers, winbash build scripts, and git (plus cross-platform IDE). Madness! A friend of mine even has to endure web dev via Citrix. Have fun doing CSS transitions on Edge in a VM in a browser tab. Or attempt to get Teams meeting invites from Outlook-in-VM out to your native or web Teams client on your local machine (as required, you know, for audio/mic). It's terrifyingly stupid.

I had one project I was working on for a client where they thought they were super important and confidential (NVM they were neglecting all of the real security best practices). One of the things they decided is that as part of the project everyone would have to move a VDI solution in a different network and then everyone would have to do all their work from this network, in the name of security. The problem was I was only interacting with the cloud side of this project, working with the actual cloud public API, so everything I was working on was already public facing and didn't care about network concerns.

I ended up writing an email to the decision markers in the project outlining why "1) I didn't want a VDI 2) A VDI would take me forever to setup properly for my workflow anyway* 3) It would do nothing for the project's security to have me working on a VDI and 4) It would be a pain in the but trying to work on a VDI.

I was told that my objections were valid and true but I needed to work on a VDI anyway. It was at that moment I stopped caring about the project.

I did end up winning in the end however as when I finally agreed to do it (not my money the company was setting on fire) they asked me to send a list of what I'd need my VDI to be setup like. So I sent them a list of everything I'd need to mimic my current working environment and what I was doing now, including a Linux OS (The VDI solution was Windows only) with X11 and i3 setup. A full Doom Emacs configuration, special developer libraries and tooling, and several other items. With a long list of alternatives I would need installed if my first choice wasn't available, with specific versions, etc.

I sent off that email and they said they'd set me up with the desktop environment I need and let me know when it was finished being created. I am still waiting for that email letting me know it is setup, and happily working from my real machine until it does become available.

The moral of the story is sometimes the easiest way to thrawt management incompetence is to throw it reams of useless busy work that it will be festering away on while you can get real work done.

Malicious compliance is a great strategy when done right.

Same!

After out-of-hand saying "no" once, I had someone offer me a contract by way of a friend. I valued the friend so I met with the client. About half way through, the client's (a large-ish health org) point-of-contact asked about my internet connection because they wanted me to remote desktop into their windows-citrix-machine-thingamputer for "billing purposes"... it was an iOS(in swift) developer role.

I didn't say anything about how preposterous that was and politely declined in email after the meeting. Or so I thought. I, apparently, did so much talking with my eyebrows that it got back to my friend... we broke out laughing when I explained what caused be to be "visually rude" (whatever that means).

When there is one red flag, there are bound to be others.

> Unless they mean it'll instantly show a loading screen, I'll bet it doesn't.

You absolutely can. I am not sure what tech Microsoft is using, but a similar experience can be had with VMware Horizon VDI product with something called Instant Clone, which is essentially forking an already booted up Desktop VM. On each esxi hypervisor host, you would have one of these parent/seed VMs that is booted up and ready in a frozen state, and when a new desktop VM is needed, it's created from the seed/parent using copy-on-write memory/disk and runtime state less than a second.

They don’t start out that way.

Usually the issue with VDI startup times these days is the layers on layers of group policy and agents that need to execute. You don’t notice on PCs because most folks lock their devices instead of rebooting.

Latency like that sounds like a problem in the company network stack somewhere… UK to Netherlands isn’t exactly a high latency wan link.

They can keep it hot in memory similar to serverless. If everyone uses the same base copy of Windows, then they can easily fork from one parent copy. A copy on write network filesystem + some Windows Registry and Explorer tricks can take care of the remaining 20%.

> then they can easily fork from one parent copy.

Ironically windows does not implement fork()

AFAIK Windows kernel is advanced enough to implement fork if necessary. WSL 1 uses it.

Think at hypervisor level. Most VM solutions support some form of memory deduplication. VM level cloning wouldn't be too far from that model IMO.

This is exactly what VMware Horizon VDI does to create desktop VMs very quickly, called Instant Clone (aka VM Fork), and has been around for several years now.

With that, you don't even need to do deduplication after the fact, it comes for free. When a new VM is forked, it's memory and disk are copy-on-write from the source/parent VM.

Windows NT was POSIX compatible.

The ZwCreateProcess kernel api can do fork() semantic.

It is possible. But if achieving a reasonable performance was doable, they wouldn't need wsl2.

that's a non-sequitur. windows not being Linux at the same time that a Mac was also Unix is what created WSL 1, and the limitations of WSL 1 led to WSL 2.

Neither does GitHub and yet...

Windows Sandbox on Windows 10 is pretty quick to boot, and uses a base copy of Windows. I wouldn't be surprised if that's similar to whatever solution they've created for Cloud PC.

I've recently used it to validate/automate whether one of our graphics tool using d3d12 would work, as it had to be deployed on some cloud machine. With small amount of batch file coding and some flags to enable file sharing and DirectX I was able to do it (for some reason, could not make it automatically start the app, but since it was for semi-automatic testing, it was okay - if I tried more, I'm sure there would've been a way).

Same here, but Microsoft, having source code all the way to the client, probably can do better than third-party providers.

And the latency, I completely understand you. You learn to live with it after a while, sometimes you don't even notice it.

> Same here, but Microsoft, having source code all the way to the client

That right there is the reason they are doing this. It's not about customers or what people want it's about keeping control of the whole stack, and the power that comes with that.

Not sure if you're aware, partners have access to Windows source. They also get to review and give feedback on upcoming changes.

Citrix is a partner, so is VMWare afaik.

> 100/20 FTTC connection at home

Same. Now the client has me working in a Amazon Workspace w/ Windows 10. Typing and screen update latency is terrible, I have to run it down scaled on my 4K screen to make it bearable, it disconnects or shuts down due to inactivity and reconnect/boot reminds me of that time I booted Windows 7 on a Dell Mini 9 Netbook (probably an Intel Atom w/ 2GB of RAM) and worst of all, my happy place tool chain is back in macOS or Linux.

> the latency when typing drove me fucking NUTS!

I had the exact same experience.

Back at my old company, all of our development was done on a secure Sungrid server farm that could only be accessed from the VDI environment. The way my coworker initially showed me how to do it, the terminal was sent to an X session on the VDI, and yet everyone at the company was OK with this. I explained how typing is something that you do tens to hundreds of thousands of times per day, and that a few frames of latency every single key press really fatigues you mentally. The sungrid farm didn't support SSH (for load balancing purposes,) so I eventually had to deepdive into the spec for how to access via terminal, and wrote a script.

But that's not the worst part as I was interviewing around at other companies, I asked them how their computing is done, and you wouldn't believe how many companies have their development environment set up this exact way! These are all major semiconductor companies in Silicon Valley too, not some "IT as an afterthought" industry!

It's refreshing to see that VDIs are about as popular with developers on HN as Google Stadia is with hardcore gamers.

My windows 10 pc goes from off to login screen in under 5 seconds and opening a remote session to my Amazon windows server takes about the same. So I’m sure that instant on experience isn’t too far off. I agree the latency will be brutal and even with a gigabit connection typing on my Amazon windows server is irritating.

> Unless they mean it'll instantly show a loading screen, I'll bet it doesn't.

with some 50Gbps Ethernet one should be able to quickly load a GB or two os ram image quickly. basically, un-hobernate, with decent infra.

more sophisticated options are having some kind of generic os image in ram one can spawn an instance of, load with local per-user configurations quickly.

being able to get an os going quickly is not a new trick.

I used to think Citrix had a lot of latency, when I was working remotely for a particular client, but then one day I went to the office and used the physical PC I'd been using remotely, and it turns out that Windows just sucks that hard. Sometimes it really does just take like half a second for a context menu to appear. Ghastly performance all around, and it was a hot rod of a PC, too.

This comes down to crap VDI implementation. I had a Citrix box at a previous gig, and I could stream YouTube from New York to Florida with very little latency (watchable for the PowerPoint decks I needed to view). If there’s not a GPU connected to the VM (or a slice of one via SR-IOV) you’re gonna have a bad time.

I'd be skeptical too if they hadn't made it work with XCloud.

Just use TRAMP/VSCode remote.

Reply on news.ycombinator.com