3 ms·
I've been playing around with pure-canvas web apps that are set up as follows: Client -> Server: Raw event stream (keyboard/mouse/touch/resize events) Serv
by bob1029 6y ago
I've been playing around with pure-canvas web apps that are set up as follows:
Client -> Server: Raw event stream (keyboard/mouse/touch/resize events)
Server -> Client: Canvas draw batch stream
The server provides a small javascript shim that bootstraps a websocket & subscribes to all required event sources. It also subscribes to the server issued batch events and has logic to dispatch draw commands to the canvas element.
On the server, I use LMAX Disruptor to aggregate the client events and process them in micro batches (the size of which are determined dynamically based on backpressure). This results in an incredibly low-latency/low-jitter UI. Client events are passed around as readonly structs, so very little allocation or GC is involved throughout (.NET5/C# codebase). Processing is concluded with async/parallel dispatch of the appropriate client-side draw batches as they are ready to be issued. Not all client events result in a draw, and not all client events result in draws on the same client. There are some synchronous concerns between clients in my application, so having a single fast thread allows for lock-free processing of all events at the same time.
The only caveat I have encountered is the latency constraint. Going over ~100ms makes this sort of interface feel like crap. For LAN/localhost, this approach is effectively instantaneous.
- kizer 6y agoI’ve considered making something similar, but I’ve thought: what’s the advantage? Just more computation for the server that could be done client side. Why not just have the app on the client. I’ve dubbed this the “cloud gaming” approach, lol.