Et voici le point de vue de Matt Mackall, développeur de Mercurial :
« fsync is a) immensely slow (can cause system-wide filesystem stalls for
many seconds) and b) doesn't guarantee file integrity on modern hardware
(many drives only guarantee that data has made it to their onboard
caches).
Because fsync has such a negative effect on average I/O bandwidth, it
can actually increase the odds of hardware corruption by widening the
corruption window. It's not at all clear that fsync is desirable. »
[^] # Re: Le post de Matthew Garett est aussi très intéressant
Posté par Antoine . En réponse au journal Don’t fear the fsync!. Évalué à 8.
« fsync is a) immensely slow (can cause system-wide filesystem stalls for
many seconds) and b) doesn't guarantee file integrity on modern hardware
(many drives only guarantee that data has made it to their onboard
caches).
Because fsync has such a negative effect on average I/O bandwidth, it
can actually increase the odds of hardware corruption by widening the
corruption window. It's not at all clear that fsync is desirable. »
http://www.selenic.com/pipermail/mercurial/2009-March/024582(...)