Re: [PATCH 0/4] RCU: various merge candidates

[Dipankar Sarma <dipankar@in.ibm.com>]

On Mon, Aug 28, 2006 at 06:33:09PM +0200, Arjan van de Ven wrote:
> On Mon, 2006-08-28 at 21:59 +0530, Dipankar Sarma wrote:
> > On Mon, Aug 28, 2006 at 06:15:48PM +0200, Arjan van de Ven wrote:
> > > On Mon, 2006-08-28 at 21:38 +0530, Dipankar Sarma wrote:
> > Hi Arjan,
> > 
> > See this for a background - http://lwn.net/Articles/129511/
> > 
> > Primarily, rcupreempt allows read-side critical sections to
> > be preempted unline classic RCU currently in mainline. It is
> > also a bit more aggressive in terms of grace periods by counting
> > the number of readers as opposed to periodic checks in classic
> > RCU.
> > 
> 
> hi,
> 
> thanks for the explenation, this for sure explains one half of the
> equation; the other half is ... "why do we not always want this"?

It comes with read-side overheads for keeping track
of critical sections and we need to carefully
check its impact on performance over a more wide variety
of workload before deciding to switch the default.

See table 2 of page 10 in this paper -

http://www.rdrop.com/users/paulmck/RCU/OLSrtRCU.2006.08.11a.pdf

Thanks
Dipankar