I'm interested in the performance characteristics of SignalR at scale ... particularly, how it behaves at the fringes of capability. When a server is at capacity, what happens?

• Does it drop messages?
• Do some clients not get notified?
• Are messages queued until all are delivered?
• And if so, will the queue eventually overflow and crash the server?

I ask because conducting such a test myself would be impractical, and I'm hoping someone could point me to documentation speaking to this ... or perhaps someone could comment that has seen how SignalR behaves at scale.

Thanks!

note: I'm familiar with this other stackoverflow question on the stability and scalability of SignalR. But I believe my question is asking a slightly different question in that I'm not concerned with the theoretical scaling limits, I want to know how it behaves when it reaches the limits ... so I know what to be on the lookout for.