Re: [Tracker] [Strigi-devel] Indexers comparison

[Jamie McCracken <jamiemcc blueyonder co uk>]

jos poortvliet wrote:
> Op woensdag 17 januari 2007, schreef Michal Pryc:
> > Hello,
> >
> > I am sending *small* comparison of the indexers. This might start some
> > discussion on many aspects of the indexing tools, to get better ones.
> >
> > In the future it would be nice to see common search query standard, a
> > lot of work have been done on it so far, but also common plugins for
> > indexing would increase number of supported formats significantly, so we
> > have plenty of work to do :-)
> >
> > On the xdg Mailing List At lists.freedesktop.org there are many
> > discussions on the search query standard, I would recommend to join them
> > if you are interested in the subject.
>
> what i really think is wrong in this article is the view on cpu usage. the 
> author (you, of course) seems to state it is bad to have a high cpu usage.
>
> why do you want low cpu usage? what is it good for to only use a processor for 
> 70%?
>
> unless, of course, some other process (the user) wants access. at that moment, 
> you must ensure the user is not bothered by the indexing. that's why the gods 
> (kernel developers) invented priority support in the kernel. just lower 
> priority (assigning a positive nice value).
>
> beagle is even misleading the 2.6.x kernel in making it believe it is a 
> time-critical process, and not hogging resources (as it is, when indexing). 
>
> the kernel will notice it doesn't use it's timeslices to the max, and increase 
> it's priority dynamically, thus allowing the beagle daemon to interupt user 
> processes! so beagle's way to lower cpu usage is actually making the 
> situation worse.
>
> all these daemons should implement support for the relatively new 'batch' 
> priority in the linux kernel. a thread or process running at 'batch' priority 
> can have all the cpu processing power it wants, UNLESS ANY OTHER PROCESS 
> requests access to the cpu. thus, such a process finishes its work as quickly 
> as possible (you'd want that) but NEVER EVER interupts a user's other 
> processes and threads. i guess solaris has such a priority as well, and if it 
> doesn't, just lower the priority (high nice value in the solaris kernel as 
> wel?), that's a start.

Im not so sure - the problem is on some machines using a sustained 100% 
cpu can cause laptops to heat up quickly, become more noisy (fans), 
drain battery life faster etc

The ultimate goal here is to remain as unobtrusive as possible while 
still indexing at a reasonably fast rate.

We do support an optional turbo mode in tracker where the goal is to 
index as fast as possible (Beagle AFAIK also has something similar here) 
so our view is to carry on down that track and support both.

--
Mr Jamie McCracken
http://jamiemcc.livejournal.com/