To Sync or Not to Sync
One reader's comments in an article I wrote about Entity Framework's async programming turned into an interesting discussion on the role of asynchronous programming in the modern world (interesting for me, at any rate). Back in the day, I used to tell developers that the surest way to create a bug they'd never be able to track down was to write what we used to call "multi-threaded applications." I gave seminars on designing multi-threaded application where, perversely, I spent the first five minutes explaining why you shouldn't do multi-threading unless you absolutely had to. And then, of course, I'd go home and write multi-threaded applications for my clients: do as I say, not as I do.
Obviously, multi-core processors and the new async tools in the .NET Framework have changed the environment tremendously. But I discovered during the discussion that I still think of asynchronous programming as something you do when you have specific issues in your application that only asynchronous programming will solve. In my mental model, I would write my code synchronously, see where I had responsiveness issues, and then rewrite those parts of the application to use asynchronous code (which, on occasion, could trigger some architectural changes to the application). Only if I could see the responsiveness problems in advance would I start by writing asynchronous code. But I was always doing it as an alternative to my default mode: writing synchronous code.
The commenters challenged that, effectively saying that (in many cases) asynchronous programming should be the developer's default choice. As one reader pointed out, a blocked thread can take up to a megabyte of memory while it's idling. Integrating async programming can eliminate that loss.
Of course, there is a question of whether you care about that megabyte: for about $10,000, I can get a server with 256 gigabytes of RAM -- that's over a quarter of a million of those megabytes that we were worrying about saving. The fully loaded chargeback for a computer programmer would swallow that ten grand in a couple of days; so if the programmer is spending much "extra" time to save the odd megabyte, it doesn't make fiscal sense.
But here's the point: If the costs of writing the code asynchronously from the start is "trivial" (to quote another reader), shouldn't I be writing asynchronously from the beginning? You wouldn't need to buy the server, and while the incremental cost for the developer time in writing async from the start might not be zero, it could easily be negligible.
It's a powerful argument. I don't think I'm there yet (I still see the async tools as making it easier to do asynchronous programming
when I need to do it), but I may be coming around. I still worry about those bugs you'll never track down, though. The exception EF raises when executing multiple asynchronous processes on the same context seem to me to be the kind of problem that wouldn't occur during development, but would raise its ugly head in production, for example. But I may be worrying unnecessarily.
Posted by Peter Vogel on 07/08/2014