News Microsoft only planned for 200,000 users for Flight Simulator launch, admits drastic underestimate led to database being overwhelmed

NinoPino

Respectable
May 26, 2022
482
298
2,060
Microsoft...,
they develop an enterprise grade SQL Server, the most used OS in the world, manage and deploy one of the top three cloud infrastructures, produce a game console, have a cloud gaming service, and they are not able to manage the launch of a game like this ? 200000 users makes their systems to crash ?
It is really scary, to think about Windows development and reliability after this.
 
  • Like
Reactions: scottsoapbox

JamesJones44

Reputable
Jan 22, 2021
851
779
5,760
200,000 seems insanely low considering the hyping that has been done and the fact that it's part of Xbox/PC Game Pass. I would think the number of pre-installs was higher than 200k. Even if only 1% of Game Pass users installed FS2024 that would still be 340,000 users (based on the 34 million subscribers it has).
 
Aug 31, 2024
2
0
10
im curious if game still uses 81GB an hr of flying as if it does the typical US Datacap wont even last 15hrs of flying. (as most caps are 1TB-1.2TB)

That was an unoptimized, alpha version of the game running at a worst case scenario. Typically, from what I have read, expect to use 4-5 GB per hour…that is, when its actually working.;)
 

TDRare

Distinguished
Jun 1, 2016
6
2
18,515
In a YouTube video, Microsoft revealed that only stress-tested the MSFS 2024 database with 200,000 virtual users. This drastic underestimate led to the database being overwhelmed and crashing on launch day.

Microsoft only planned for 200,000 users for Flight Simulator launch, admits drastic underestimate led to database being overwhelmed : Read more
And yet, they obviously had way more pre-orders than that if more players were available to login. People who pre-order intend to play at launch - why is that a difficult concept fro game studios? They should be watching pre-order numbers and continuously planning server capacity for at least 25% more than that. Worst thing you can do is provide a lousy launch experience. Can you imagine paying $260 USD for Aviator pre-order, than failing to launch or receiving a server queue to wait? We should expect better.
 

doughillman

Distinguished
Feb 13, 2012
57
67
18,610
That was an unoptimized, alpha version of the game running at a worst case scenario. Typically, from what I have read, expect to use 4-5 GB per hour…that is, when its actually working.;)

I've used about 3GB per hour today. That said, probably about half of my flying has been out of my home airport and in the surrounding area, so a lot of the terrain I've flown over today has likely only been downloaded once and remained in cache.
 
Lots of people are playing it on game pass which makes microsofts low player estimate strange. You have people that would never pay money for the game but will try it out when they get it included for basically no extra cost.

True. Going by various Steam charts (Steamcharts and Steamdb) MFS 40th Anniversary had under 62,000 concurrent players at any given time, so one could see how a 200,000 stress test could be considered adequate, but with the drastically reduced download size and huge promoting of Gamepass Microsoft has been doing...
 

DS426

Upstanding
May 15, 2024
254
188
360
One of the largest tech mega corps in the world and they botch it this bad... my only explanation is that Microsoft is just greedy at the end of the day -- too greedy to provide, oh I don't know, even average customer satisfaction and service for their markets.

Lol game launches... glad I'm not a part of these frenzies. I pre-purchased Farm Sim 2025 and sure enough on day 1, it's crap from both performance problems and plenty of game bugs, so waiting two to four weeks for patching to improve things. No biggie -- got plenty of other games to play and more important, real life chores, right!?
 

abufrejoval

Reputable
Jun 19, 2020
582
421
5,260
First impressions:

Caveat: I'm no pilot, nor would I want to be one. That's one thing that having used every version of FlightSimulator since the first on my (last ever) Apple ][ taught me: most of the time it's just too boring.

Consequence: I'm mostly doing it for the tech interest and because I like the idea of armchair travelling.

Problem: M$ has exaggerated their promises of world view realism to the point where they should just be sued to put up or shut up. This time around they actually promise a "digital twin" and that is quite simply a blatant and inexcusable lie.

What I used for testing:
For the armchair travelling I need bandwidth on the eyes. So it's either big screen or VR.

My main screens are two 43" 4k monitors:
  • the original "work-only" is a 60Hz IPS without anything fancy and now sits to the right as an optional screen e.g. for monitoring
  • the newer "game-ready" is a 144Hz VA with HDR1000 and often runs as sole primary
They run through a mess of partially cascaded KVMs, but the primary is a dual DP, which means that anything critical (really hi-res, Hz and bit-depth) only has one KVM to go through.

I also used three systems:
  1. Ryzen 9 7950X3D with 96GB ECC DDR5-5600 and RTX 4090 @4k
  2. Ryzen 9 5950X with 128GB ECC DDR4-3200 and RTX 4070 @4k
  3. Ryzen 7 7435HS with 64GB DDR5-5200 and RTX 4060m (laptop) at 1920x1024
all with WD850X NVMe storage. 1+2 share a 10Gbase-T network with a Gbit Internet uplink, 3 uses a 2.5 Gbit NIC on the same network.

VR was mostly a Quest 3 via SteamLink connected via a WIFI7 access point on the 10Gbit network and without other traffic. An HP Reverb didn't do any better, so I didn't explore further.

Personal best changes vs. FS 2020:
  • changing any graphics settings, resolutions, quality, refresh rates, features like DLSS etc. happen immediately without a restart and don't much affect the 2nd monitor! That's really cool especially when there is so many of these dials to turn these days and you'll all remember how many times you had to restart things in the olden days...
That makes it really easy to test the various settings and see both the quality and performance impact.​
There is plenty of other changes, but they don't change my main gripe:

What is sold as a "digital twin" is so badly faked, it's hard to stomach.

The main problem is that Microsoft doesn't own the data it would need for a digital twin. They use Bing Maps and Bin Maps is very bad quality data. It's first of all very variable, so in a lot of places they have very low resolutions or they might have different resolutions from different time periods.

I've noticed that when I flew (using FS2020) over one of my employer's office buildings in Lyon (and in fact all of Lyon, which isn't exactly a small hamlet, seems to share this) I'd do time travel, when I'd get closer. From a distance, it would be modern recent data, but once I got close, it would revert to Lyon as it was 20-30 years ago, office buildings would turn into old factories, etc.

And it's the same on Bing. So just have a look there, if you don't think that's very good, current or realistic, it won't get any better with FS2020 or 2024, only the degree of hallucination will increase. And while it used to be generated on the client side in older editions of FS, today it seems to be generated and rendered on the server side and is then streamed as "ground truth" over the Internet, giving you a permanent fake twin or a bizarre near empty wasteland when disconnected.

My standard test flight is flying from Frankfurt (one of the bigger airports around here) to an army airfied in Wiesbaden, which has me pass right over my house, and the towns and villages I know best.

Yet it's very hard to recognise any of that, because for lack of data FS (2020+2024) invent the very same buildings, barns, vegetation etc. which have very little to do with what's actually there. It invents a very generic set of buildings with a fixed style that resemble the house and hotel pieces from Monopoly in form, and are then painted or skinned in a sort of rustic 1900's style. And since it completely lacks elevation data everwhere it doesn't have "specially augmented" data, mansions turn into huts and barns into factories etc.

And then there is river and road crossings, sometimes with traffic. I've never seen so many bridges fallen into rivers, sunken fleets of pleasure boats, and fleets of cars zipping into and through each other on 4 lane crossings: the carnage and visible catastrophies completely destroy the realism of the fake reality they've developed and copies for decades now with zero AI improvements.

Once you go a few thousand feet into the air, none of this may matter any more, but white-on-white clouds for 8 hours or more on a transatlantic crossing aren't what I aim for when lauching FS.

Where there is detailed 3D data, it's rendered at somewhere between DX7 and DX9 levels of visual quality, the same goes for the plane insides: the amount reuse of old data and code is giant, yet I've always had to pay for a full new version with small but overhyped incremental improvements.

The advantage of using low-tech graphics is performance. Obviously a photo-realistic digital twin is impossible, and the vastly improved rendering of badly faked data won't improve user experience.

So where past editions struggled on higher-resolutions, because they were mostly single-threaded and used serializing graphics APIs, FS2024 runs rather well even at 4k on the 5950X even at ultra settings: ~45-70FPS depending on DLSS aren't bad for a mere RTX4070.

On the RTX 4090/7950X3D even FS2020 finally did sort of ok, and FS2024 barely ever dips below the 144Hz I've set as a maximum on the Nvidia panel on Ultra settings.

Most importantly, even at the lower frame rates on the weaker 4070, the world moves smoothly and that is extremely important for the illusion of flight.

And that's the other giant gripe I have with FS2024: the world stutters and stops where it counts most, in virtual reality, which kills any illusion of flying.

So here is what happens: as soon as I switch to VR via Ctrl-Tab the world stutters the view doesn't.

The inside view of the plane and the reaction to any head-movement is buttery smooth at the 90Hz the Quest runs on (good, no bucket required), but everything beyond the glass panes, or simply the outside world stutters as if it was swapping pages to hard disks.

It's very variable, includes hangs of several hundred milliseconds now and then and simply feels like running the sim at 10-20Hz (again with the "inside" VR view remaining completely fluid).

Now the very same GPU has to render both, the headset itself may be very smart and quite 3D capable (in the case of the Quest 3, HP Reverb is reall just displays and sensors), but here it's doing little more then providing sensor data to the PC and then streaming the RTX4090 generated data back.

And it has to prioritize the inside view to avoid nausea. But from what I can tell the outside view isn't really taxing it.

I'm using HWinfo, etc. to monitor the load, and there is plenty of horsepower left.

I've also drastically reduced the (monitor) screen resolution and quality to lighten the GPU load, but it makes zero difference. Even the quality settings for the VR side of things don't have any impact, quality levels, DLSS support etc. dymanic scene updated frequency (30/60/90Hz tested) don't impact the stuttering in any significant manner.

What's really frightening is that this hasn't changed from FS2020 to FS2024: both stutter in VR and quite indepement of the headset (The HP Reverb has a very similar resolution to the Quest 3 but uses a DP cable instead of Wifi).

And again the quality of the inside view proves that it's not a Wifi issue, as do all the other VR games.

It's an issue with how VR affects the simulation core. And that's a real shame, because that simulation core is doing surprisingly well on a flat screen, even when FPS fall to 20-40Hz on an otherwise overtaxed GPU e.g. without DLSS.

I've even run FS2024 in a "windows" stretched across both screens at an effective 8x4k resolution on the RTX 4070 and only had to lower from "ultra" to "high" to get fluid world and screen refreshes...

The other world data issue:

I also had to try several times and over longer periods to get FS2024 going and into my test flight.

Install and startup are atrocious and reflect very badly on Microsoft, not that I've ever held them in very high regard.

But what really made me uneasy is that the whole flying experience with all its defective world data, becomes a bizarre pixmap nightmare when Microsoft's servers are overloaded and cannot provide world data.

Even though I had increased the rolling cache to 64GB immediately and didn't do large excursions, when competing traffic was high, visual quality would drop off even further, resulting in a very abstract outside world. The same flight a few hours later with evidently less competing traffic had at least the "high-precision fake digital twin" data (btw. I didn't notice FS using the 64/96/128GB of RAM the machines have in any significant way).

I'd typically get a "low bandwidth" warning then, but it sure wasn't my side of the network: my fiber Gbit Internet uplink is really constantly near theoretical max as Steam updates or a quick Linux ISO image download torrent will verify.

So it's clearly an Azure server side issue, and nothing constant.

The uneasy part is that even with the evidently badly faked world data, it's a lot of it and serving it out to customers who currently can't even download fixed sections for offline use, is a significant effort, which Microsoft will cease to support, far before I'm likely to want to stop flying.

I've bought games years ago I plan to play during hopefully decades of retirement and here what was clearly a game purchase hints at a limited time service, which for FS 2020 might actually be terminated before they've even worked out the bugs.

Yeah, my adventure isn't flying. That's too boring once you're up there, and it doesn't offer the digital twin armchair travelling that Microsoft is selling.

But at least testing, profiling, etc. was two days of adventure so far. Yet somehow I think like Microsoft should be paying me, not the other way around!
 
Last edited:
  • Like
Reactions: Bikki and P.Amini

Sangheili112

Commendable
Aug 31, 2022
8
0
1,510
im curious if game still uses 81GB an hr of flying as if it does the typical US Datacap wont even last 15hrs of flying. (as most caps are 1TB-1.2TB)

It depends on graphics settings and location like grand canyon or cities vs flat land like the desert

I think 4k or it seems is streaming but low quality might be downloaded
 

abufrejoval

Reputable
Jun 19, 2020
582
421
5,260
im curious if game still uses 81GB an hr of flying as if it does the typical US Datacap wont even last 15hrs of flying. (as most caps are 1TB-1.2TB)
The older versions of FS had very little and very rough terain data and invented quite a lot of stuff, especially anything animated.

FS2020 evidently started with trying to "improve" things towards a "digital twin" experience using Bing data, which is also faked, but on the server side. Google does that too, but in a far more consistent manner, with much better base data, but at a render quality that clearly wouldn't match a flight simulator's visual requirements.

Anyhow, that leads me to believe that most of that data comes basically in a metafile format or what would look like log of API traces. So mostly vectors or rather triangles with a texture for buildings and a bit of procedural metadata perhaps for vegetation which is "grown in place". I can easily imagine that roads, rivers, lakes and some other standard terrain building blocks have their own non-bitmap description formats for small size and resolution independence.

It explains the fact that badly faked buildings in cases where augmented 3D has not been purchased or generated by Microsoft, actually remain the same when you fly over them later. In fact my neighborhood has been faked very badly but with complete consistency across FS2020 to FS2024, because it comes from Bing Maps.

It makes for pretty similar and (important!) predictable data volumes transmitted for areas with low-res data without elevation and areas where Microsoft has detailed 3D models and it also makes it much easier to scale (distance and user screen resolution are after all very dynamic) to the user's viewport. I haven't tried to check if they manage distinct resolution bitmaps, but the whole graphics technology there is a lot like really old DX7 or similar stuff from more than 20 years ago to manage data volumes.

Even if the visual quality achieveable is far below what Microsoft's marketing has consumers expect, I appreciate the technical challenges and the work that's been put into trying to make the impossible somehow manageable.

The algorithms and functional infrastructure (operational scaling issues aside) for a much better and more accurate digital twin seem to be there, but without Microsoft spending money on better quality data (which is abundand today, but not cheap) consumers won't get what they thought they bought.

Land lines in Europe don't have capacity limits so I haven't looked very closely. But server bandwidth availability is currently a big issue so I've looked at least a little bit at what the rolling cache size should be to avoid flying through a featureless moonscape if online world data isn't available.

Unfortunately it doesn't give you any immediate feedback on how effective your rolling cache is, but flying around the metro area here for a few hours only made for a few gigabytes of world data downloaded.

I'd hazard that data volumes are as modest as the visuals, but I'm sure that more data on this will arrive soon.
 
  • Like
Reactions: P.Amini