Re: New module proposal: tracker

[Alexander Larsson <alexl redhat com>]

On Fri, 2009-11-06 at 10:15 +0000, Martyn Russell wrote:
> On 06/11/09 09:41, Alexander Larsson wrote:
> >
> > I think the inotify issue is only part of the problem. Whats needed is
> > generic work on Tracker and/or the kernel so that its possible to run it
> > while not affecting system performance as much. This kind of lowlevel
> > work is hard, but I see it as really important, because as soon as there
> > is any disturbance to general system performance when Tracker-indexer
> > runs a majority of people will remove it because they won't think its
> > indispensible (since they haven't even started using it much). This
> > leads to a Moment 22 thing where people not using it leads to
> > applications not supporting it and thus it not having interesting app
> > integration and thus leading to less users.
> 
> We have spent a lot of time over the past 6 months improving this. I 
> would say 0.7 is streets ahead of what 0.6 used to be like in this 
> regard. I think to some extent it depends on your data set but I don't 
> notice tracker 0.7 working in the background on my home machine or 
> laptop (other than the icon that updates of course) and I tend to have 
> it watching my source directories while I work.
> 
> There might be more that we can do in this regard, but a lot of effort 
> has been put in this area because on a device like the n900, you can't 
> afford to hog CPU when on a phone call (for example).

Thats cool. Although I don't think CPU load per-se is the main problem.
CPU scheduling is pretty easy to control such that a process only runs
when nothing else runs. The main problem is i/o costs (increased amount
of seeks causing degradation of application i/o) and general VM
behaviour (filling buffer caches, bumping out other apps from memory,
etc). These things are much much harder to control and measure.