[2327] in athena10
Re: Decoupling Firefox from AFS slowness
daemon@ATHENA.MIT.EDU (Evan Broder)
Sat Apr 18 23:55:07 2009
Message-ID: <49EAA07D.1020509@mit.edu>
Date: Sat, 18 Apr 2009 23:54:37 -0400
From: Evan Broder <broder@MIT.EDU>
MIME-Version: 1.0
To: Kenneth Arnold <kcarnold@mit.edu>
CC: Greg Hudson <ghudson@mit.edu>, debathena@mit.edu
In-Reply-To: <d4d4976a0904182041x231bc82docb75a215754702c3@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit
Kenneth Arnold wrote:
> On Sat, Apr 18, 2009 at 7:06 PM, Greg Hudson <ghudson@mit.edu
> <mailto:ghudson@mit.edu>> wrote:
>
> Is git really buying you anything here? It's designed to version
> source
> files, not sqlite databases, so I foresee a lot of potential for
> impedance mismatch.
>
>
> No it's not. I vaguely like the idea of my bookmarks being tracked by
> git, but for just about everything else it doesn't make sense.
>
> Much-improved script attached, converted to use rsync instead. It
> should suffice to mv .mozilla .mozilla-master before trying it.
>
> It still leaves sleep processes laying around after Firefox exits. Any
> ideas why?
>
> Okay, now I'm /actually/ stepping away from this; there are more
> important things to do! :)
>
> -Ken
Ugh...regardless of whether Firefox is slow working out of AFS or not,
this approach worries me, because you're basically turning this into a
caching problem. How do you make something reasonable happen if firefox
gets killed out from under you? Or even worse, SIGKILLed? Does the user
just lose on any changes they've made to their profile? That seems
unacceptable to me.
If we want to make Firefox run faster, let's do it in a way that doesn't
displace the user's profile. Maybe there are more settings that we can
frob to make it perform better?
- Evan
