DragonFly BSD
DragonFly kernel List (threaded) for 2003-07
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: token passing & IPIs


From: Hiten Pandya <hmp@xxxxxxxxxxx>
Date: Fri, 18 Jul 2003 04:14:45 -0700

On Thu, Jul 17, 2003 at 07:20:00PM -0700, Matthew Dillon wrote:
> 
> :
> :	Are you planning to use a Read-Copy-Update mechanism for
> :	file system lookups?  This could provide scalable performance
> :	according to the theory to which it works.
> :
> :	Just my 2 cents. :-)
> :
> :	Cheers.
> :	
> :		-- Hiten (hmp@xxxxxxxxxxx)
> 
>     I don't know enough about the mechanism to have an opinion one way or
>     the other.  Read-Copy-Update sounds suspiciously like an MESI model.

	It's the lock-less SMP stuff that Linux is using.  The idea
	originated from IBM's Paul McKenney; and Rusty Russel of Linux
	Camp actual made it work for Linux 2.4.x and later versions [2]

	It will be a good idea to do lock-less file system lookups,
	since it is a lookup, there is no write to it, thus we can avoid
	locks as much as we can.  So can we for device state tables, and
	such.

	They have seen a significant scalability gain according to the
	results, and the basic theory behind how
	Read-Copy-Update/Lock-Less works [1].

	[1] - http://lse.sourceforge.net/locking/dcache/dcache.html
	      http://www.rdrop.com/users/paulmck/rclock/intro/rclock_intro.html
	      
	[2] - http://lse.sourceforge.net/locking/rcupdate.html

	Cheers.
	
		-- Hiten (hmp@xxxxxxxxxxx)



[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]