Comments on: More details on the fsync solution

By: Front-end Performance in Firefox 3.5 and Beyond « dietrich

Front-end Performance in Firefox 3.5 and Beyond « dietrich — Tue, 05 May 2009 20:42:15 +0000

[...] so that it no longer needs to write to the database every time you visit a page, a project well chronicled on Shawn’s blog. Spurred on by one of Shawn’s performance wins, Ed Lee was able to [...]

By: Shawn Wilsher

Shawn Wilsher — Mon, 20 Oct 2008 01:37:55 +0000

@asdf

That doesn’t actually provide us with everything we need. We’ve looked into it.

By: asdf

asdf — Sat, 18 Oct 2008 22:19:53 +0000

On Vista you guys should use the new transactional NTFS API so you don’t have to do this kind of fsync ping-ponging nonsense to try to emulate transactions poorly.

Ditto for unixish systems that use manly file systems like ZFS.

By: Issues Issues…

Issues Issues… — Wed, 06 Aug 2008 07:01:12 +0000

[...] Calendar More details on the fsync solution [...]

By: Shawn Wilsher

Shawn Wilsher — Tue, 05 Aug 2008 18:37:57 +0000

@Robert

If it were that simple, this probably would have been solved for Firefox 3. Sadly, for ext3 (and many other linux file systems) we have to work around a file system bug. In the case of SSD storage like a USB key, it’s just plain slow. We need to get these writes off the main thread anyway, so this is overall a good thing to do.

I do however agree that it’s complicated. It is, however, the least complicated solution we’ve been able to really come up with.

By: Robert O'Callahan

Robert O'Callahan — Tue, 05 Aug 2008 18:21:20 +0000

This seems pretty complicated. Is it really impossible to just make SQLite address the problem internally?