Re: [Boost-users] boost asio synchronous vs asynchronous operations performance

Subject: Re: [Boost-users] boost asio synchronous vs asynchronous operations performance
From: Bjorn Reese (breese_at_[hidden])
Date: 2014-03-30 10:49:49

On 03/30/2014 04:13 PM, Niall Douglas wrote:

> Just to clarify, my ~9000 cycles/op is for maximum contention i.e.
> fully loaded with eight threads all fighting it out. Is your 1000
> cycles/op for two threads only?

Yes.

>> It locks/unlocks four times per operation, which accounts for a
> total
>> of 20% of the CPU time.
>
> I'm actually surprised it's as much as that. I would have thought
> twice per operation is the minimum possible, but you have to make
> some hard design choices to get it that low. AFIO "looks funny"

It uses one lock to protect its epoll data, and three to protect the
io_service members. Don't ask me why it needs three in the latter case.

Talking about surprise, I was surprised that system::system_category()
accounted for 10% of the CPU time, but it looks like Asio wraps all
system calls in system::error_code() before using them (e.g. for
checking for would_block in non-blocking I/O.)