Re: stochastic fair queueing in the elevator [Re: [BENCHMARK] 2.4.20-ck3 / aa / rmap with contest]

[Andrea Arcangeli]

On Mon, Feb 10, 2003 at 02:29:20AM -0200, Rik van Riel wrote:
> On Sun, 9 Feb 2003, David Lang wrote:
> 
> > note that issuing a fsync should change all pending writes to 'syncronous'
> > as should writes to any partition mounted with the sync option, or writes
> > to a directory with the S flag set.
> 
> Exactly.  This is nasty with our current data structures;
> probably not something to do during the current code slush.

as I said, you can consider asychronous all requests submitted with
current->mm, this will be 90% accurate.  This whole thing is an
heuristic, if the heuristic fails the behaviour become like w/o SFQ so
no regression. Also the latency of fsync is less critical than the
latency of a read() syscall, again, this is statistically, sometime it
can be the other way around.

Andrea