Subject: Re: [boost] This AFIO review - a modest proposal
From: Niall Douglas (s_sourceforge_at_[hidden])
Date: 2015-08-24 16:11:54
On 24 Aug 2015 at 18:18, Giovanni Piero Deretta wrote:
> On 24 Aug 2015 5:15 pm, "Niall Douglas" <s_sourceforge_at_[hidden]> wrote:
> > Lock free filesystem programming is like lock free atomic programming
> "You keep using that word, I do not think it means what you think it means".
> Lock-free is a Word Of Power. Unless you implemented the whole stack (down
> to issuing disk ops) in user space or audited every supported OS kernel, I
> do not think you can claim any lock freedom property.
Thanks for pointing out there may be confusion here.
I meant lock free on the filesystem. As in, you don't hold a lock
file, and you don't take byte range locks.
AFIO is all about exposing the *memory* locking the kernel does on
your behalf to application code such that you don't have to hold lock
files or take byte range locks.
Lock files and byte range locks are slow. As soon as you use them,
you'll throttle performance to the thousands of ops per second range.
This is because filing system locks have more than braindead design,
they are breathtakingly inefficient :(
-- 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