Can A VMS Scale To 200k+ Cameras?

Are all the cameras running at QCIF and 1 FPS?

Otherwise, no.

#1, what do you mean by 'a single instance'? Does that mean all cameras recording on a single server or? I assume you don't but it appears to be interpreted that way. Can you clarify what you are looking for?

Also, I would not put this in a binary 'yes' or 'no' response when it comes to a VMS managing tens of thousands of cameras across hundreds of servers. Some VMSes will work better or worse on certain features (where scaling has a bigger impact, i.e., searching across all cameras, etc.).

Think YouTube.  It’s a video service with millions of video inputs and sources, many source types and viewed based on subscriptions and permissions by millions. 

In general, you want to see how a real 100,000 camera VMS system works.

At some (generally trivial) level, they are all 'unlimited', in the system that computers can hold millions of records (whether that is records of cameras or motion events or recording schedules, etc.).

The practical problem comes in with latency, reliability, etc.

For example, with 200,000 cameras, depending on how the system is designed and implemented, it might take 100 seconds or 10 minutes to return the results to a basic query. In such situations, it might 'work' but it would not work well enough to satisfy most users. 

Another example would be making changes. When you are dealing with so many cameras, a functionality that works manually (enter in information X one by one) may be untenable with hundreds of thousands of cameras. A vendor might say "Sure, just enter it one at a time 200,000 times", that VMS 'can' do it but pretty terrible, while another system may implement automation tactics to significantly reduce the time and error involved.

So to your question:

any VMS that can manage

I think most would argue that they 'can' but how well they actually do is hard to determine unless they have real installations at those scales. In my experience, typically really big installs have 'weird' problems that only are uncovered at scale and are generally not publicly promoted, as it is not in the interest of either the VMS nor customer.

who has 200K cameras in one physical location?

My problem is with the 'not federated' clarifier.

This is what federated architecture is designed for...

For comparison:

Genetec and Milestone 500 Camera Recorders Announced

and

Record 700 Cameras On A Single NVR - Claims Indigovision

That's an open ended question, and so an open ended "Yes" is my best response.

You seem a bit all over the place with your question and responses such as federation. I think you feel federation means beyond one building/location, but theres a lot more here. Are you asking what software can manage that many feeds on one site? What about storage, redundancy, server physical size and heat output??

Think of a federal org or a national/international company that has offices in all major city in each state (e.g. remote location) and like to manage all cameras from a single enterprise server (central site). Not just view which can be done cia PISM like applications. 

The central site is the one that needs to manage all 200k+ camera and manage user access for monitoring and investigation. 

Simple, change the way your indexing access to the stream. The VMS is irrelevant at that level.

I will suggest that large systems will need an Alexa like HMI to interface with a large database.

Think of driver licenses, do you think they type in your name? John Doe, then have to filter 1.4 million results?

There is an easier way to interface, not going to say exactly how it's done but enough information is given for at least one of the layers.

If you stay in the old frame of mind, good luck stacking a bunch of hardware.

Of course it can.  Too many DVR people think of a 'system' being a DVR.  It's not, a system is the whole shebang.  It's a bunch of smaller systems combined into single login.  Don't understand what the big deal is?  

And don't ever compare video surveillance with YouTube.  YouTube sends the same video to thousands, millions of people (streaming service).  Surveillance seldom ever works that way.  

Sounds like an interesting project. From my perspective, I would contact Milestone pre-sales in your region, and see what they say.

Indigo Vision VMS platform can do it easily as it doesn't have central server, therefor no bottleneck there. Also their NVRs are the biggest in the market (biggest one 600ch with 1.5PB storage). Unlimited number of NVRs and users in the system and beautiful thing is those instances doesn't require any license at all. Even for redundant NVRs, you just need the device itself. Only license that IV is asking for is per camera in the system + SUP. 

Main concern here would be to design proper IP network which can handle such amount of video streams. So Cisco or Juniper would be my recommendation with proper deign like access and aggregation level switches.

I work on a Federated FLIR Latitude system of nearly 20K cameras every day. Anything is possible, but as Clint quite rightly states, Network, Storage, Heat/Cooling, etc etc are major factors. Redundancy for me is a hot mirror site/DR so I'm already recording circa 40K streams!