Re: fsync/MAP_SHARED slowdown

[Rik van Riel <riel@redhat.com>]

On Thu, 20 Mar 2008 13:12:16 -0700
Brian Harring <ferringb@gmail.com> wrote:

> Back in 09/2006, rev d08b3851da41d0ee60851f2c75b118e1f7a5fc89 
> (discussion at http://lkml.org/lkml/2006/6/19/245) was commited with 
> the purpose enabling tracking of shared dirty pages.
> 
> Still trying to identify exactly why, but that specific rev results in 
> a ~3x slowdown in fsync speed for an app of ours- an isolated test 
> case demonstrating it is attached.

The kernel now includes an implicit msync for each fsync.

In short, before the change an fsync would not find all the dirty pages
of the file, because some of the dirty bits might be "hiding" in the
page tables, while the current kernel actually finds those dirty bits
and writes all dirty data to disk.

Would you rather have the fast fsync, or the one that actually writes
all the dirty data to disk? :)

-- 
All Rights Reversed