How to 'subscribe' to an event that ripples to client

I know that there are a variety of similarly posed questions here on SO and I've had a look at the ones that are suggested as matches. However, none quite manage to solve the issue that I'm facing. Basically, I'd like to know if there's a way to 'subscribe' to an event at client level in order to have a small portion of the page rendered from the controller. I know that I can use javascript setInterval() (in combination with jquery ajax) to 'poll' the controller action in order to determine if something 'new' should be renedered onto the page. However, I'm not a huge fan of polling, especially when client browsers can be left unattended and are uneccessarily polling for changes. Multiply that by the number of potential client machines that could be browsing the app and you get a feel for the scale of the issue with this approach. Now, in truth, this is EXACTLY the approach that I use to refresh certain page fragments on some ('rea开发者_StackOverflow社区d' - MANY!!) of my existing sites.

However, I'd like to know what the options are for subscribing in 'reverse' to events and having the controller 'know' about the client browser and any events that it is subscribed to and the controller then 'push' out the update to the client on a 'needs must' basis. Is this something that happens outside of the normal desktop events scenario?? is the idea even scaleable or is it just the same potential bottleneck in reverse of the method that I currently use.

Hopefully, some interesting approaches out there to this double edged sword.

The server can not contact the client. The only other option besides polling is leaving an open connection between the client and server that the server can then stream info through. This is not less resource consuming than polling.

You can write a script that detects action on the client browser and goes into a "sleep" mode after a few idle minutes during which it does not poll.

Never tried it, as it seems to me that maintaining a long-running connection is a good way to run out of worker threads¹ - but Comet seems to be what you're looking for.

1. Note that others have suggested ways around the thread per connection model.