Subject: Re: [boost] [ANN] Boost.UI - a new C++ GUI library
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2017-09-29 22:17:39
> Since you propose a Boost library, it would be a valuable exercise to
> consider how best to integrate your framework with Asio -- especially
> considering that something very like Asio is well on its way to
> becoming part of the C++ Standard.
Not sure if that was intended for me or the OP, but in AFIO's case we
cannot be in any way compatible with ASIO's io_service. On POSIX async
file and socket i/o use completely incompatible systems, and on Windows
the 99% latencies with file i/o on IOCP are an order of magnitude worse
than the alertable i/o mechanism used by AFIO.
Besides, according to empirical testing, most users will not want to use
async file i/o with anything but non-cached i/o i.e. 4Kb multiples to
4Kb aligned offsets. Async i/o with cached i/o is never better, and
usually much worse, than synchronous i/o on SSD or better storage. AFIO
v2 is designed accordingly, there isn't much async actually in there
compared to v1. Hardware and OSs have very significantly changed.
A whole new generation of storage (DAX) lands next year, it appears to
the CPU as if it were RAM, so it's basically permanently mapped into
memory and taking advantage of that 64 bit address space. AFIO performs
excellently on such storage, nanosecond scale i/o latencies. But it
completely upends how you design i/o, for example iostreams really sucks
on DAX storage, it's unbelievably inefficient. You simply cannot max out
a DAX device if you're using iostreams, not even close.
I may float AFIO for standardisation next year. C++ will need to keep up
with hardware improvements in order to stay relevant.
-- ned Productions Limited Consulting http://www.nedproductions.biz/ http://ie.linkedin.com/in/nialldouglas/
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk