PA Server and performance failure

You surely meant expensive. Now that I had my daily fill of acting the Grammar Nazi let me say that the project launched with client-server architecture being advertised as a core feature. At this point it was clear that the game was CPU- and upstream-bounded on the server side and GPU- and downstream bounded on the client side.
If your CPU is weak or your upstream bandwidth is limited do not try to host big games (or at all). If your GPU is weak or your downstream bandwidth is severely limited do not join big games. Calling for a complete architecture redesign and subsequently rebuild is plain silly.

Observers are bound to put an immense strain on a client-server engine like this. The only way I think this will ever work out is having a second server (or multiple secondary servers) close to the original one sort of mirroring the main server and multiplexing Observer data to keep the original upstream in check.

It shouldn't be so bad - there are limited players per server or whatever, right? So it's really no different from having 10 players on the server, than having 9 players and 1 observer.

I think the difference is that the observer is requesting what another player has already seen, and I don't think there's good architecture in the code to allow that to happen.

Plus all of a sudden the player is placing a lot of work on their own CPU/GPU to generate the fog of war, and remove all the units that the player has not seen

Yes it does
"Hey, where did my strike team go?"
*Magic chronocam*
"Ohhhhh".

Sure, it shouldn't matter in most cases. They are advertising 40 player games after all. But peer-to-peer lockstep is able to support many observers in addition to the actual players, so we definitely have a design trade-off here that has to be acknowledged.
And consider observers joining late into a 20 player game that is in full force with thousands upon thousands of units already present. The server and the observer client have whole of catching up to in that case. Or has that feature been dropped?

Basically other client or other observer might work for mirroring because server is perfectly know what information each client have so I guess it's possible.

If game become popular Uber can make their own CDN for this purpose because bandwidth is pretty cheap these days. And if game not become popular there won't be need in support of tons of observers. So I think observers it's very small problem.

I agree that 40 and more players in same game it's really interesting question to how it's will work in terms of bandwidth for host. For me personally 100mbps for host is real number, but something more is not as simple.

Though now something like 300mbps become new standard for servers so I want to believe game will be successful so Uber might be a sponsor for 40+ players fun at least for next 2-3 years.
And if something really bad not happen with internet after 2-3 years something like 1gbps per server won't be a big problem.

Seriously you're overestimate how much bandwidth used at moment when client "catching up" because most of bandwidth in live game it's projectiles and patrolling units. At moment when client get base state of game it's not actually need information about those entities so it's need way less bandwidth.

Though you might be interested to check this @UberWilliam post about sync for 140,000 trees:
https://forums.uberent.com/threads/units-not-appearing-after-a-crash.51817/#post-793735

But the Chronocam only scratches the itch caused by the popcorn-combat so I don't regard that as a good argument in this case.

I believe one of the primary uses of chronocam will be sharing games, and being able to use them as the savegame system.

Let's say you were playing a game, and lost miserably. You can go back to any moment of it, take control again, and fix your mistake.(Single player obviously)

Also, I understand that you have network performance issues, so do I. My games tend to be jumpy at best after the 5m mark.

But please understand this: There are a great many things that will change with this game, as it is still heavily in development, and far from finished. Just give it a bit of patience, steadily report your bugs and performance issues, as it helps getting them fixed.

Judging by what I learned lurking here for the past couple months, Uber wouldn't properly release something half-baked.

Popcorn combat? What do unruly movie theatres have anything to do with this?

Could you please cherry-quote my posts in a fashion that does not make me like you? It clouds my judgement when trying to nitpick on other people's remarks!

No.

What's a cherry-quote?

Cherry picking single elements of a quote. And yes, I can make up words all I damn well please - I am German, after all, and our language is basically built on the very notion of smashing words together left and right.
Plus, look at my manly signature - I am officially 'brutally efficient', even if it is only at compressing sentences into gibberish.

Pretty sure that's English.

There are games that do this in multiplayer so it's not so obvious.

I am trying to bring the blessing that is compounds to less fortunate languages. Kind of like the guys handing out blankets to the native American population in the days of yore.

So decimating the population and then making a half-baked attempt to allow the rest to survive the next night?

Oh cool, didn't know that. I'd assume you could do it with any game really, as If I think about it there wouldn't be anything stopping a lobby of players from joining an already saved replay game rather than creating one from scratch. It's just my mind tends to be single player focused is all.

40 players
http://pastats.com/pastats/chart?gameId=7570

Was actually pretty playable.

Based on the information we have at the moment- 40 player games should be fine...

1 v 1 with 40 spectators, perhaps not so much (although I'm sure they can probably tweak things to reduce bandwidth needed for specing soon).

more spectators does not matter at all to the clients, so why would 40 spectators not be an issue? Given a server with 1gbit bandwidth at least.

It might be an issue for the server if anything. It's fine *IF* you have 1gbit to play with. Although I'm certain that part of the issue is the information isn't being filtered for spectators properly yet. It would be fine on LAN I suppose.