Data on the network may be more high-priority than work in the queue. This is the case for example for financial applications which need to react to market data coming from the network as quickly as possible. On 27 September 2016 at 16:10, Tatsuyuki Ishi <ishitatsuyuki@gmail.com> wrote:

The current run/poll_one does polling regardless if there's task in the queue.

If there's task in the queue, polling is just a waste of system call, since there's already a job to do.

Also, my common use case is "wait and consume all work", which is not covered efficiently by current design. (run_one+poll takes two system calls)

Side note: on Windows asio performs even worse, it only retrieve one event per syscall... Not sure if it's intended.

Friendly note: please use "Reply to All" when replying to me.

_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/ mailman/listinfo.cgi/boost

On 27-09-16 17:10, Tatsuyuki Ishi wrote:

The current run/poll_one does polling regardless if there's task in the queue.

If there's task in the queue, polling is just a waste of system call, since there's already a job to do. This is not accurate. You cannot do fair scheduling unless you know the work available. Neglecting some of the work leads to possible starvation of those tasks (imagine if the task in queue posts a new task every time. Your suggestion would **never** poll at all).

Seth