Subject: Re: genfs_putpages with PGO_SYNCIO
To: None <enami@sm.sony.co.jp>
From: YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
List: tech-kern
Date: 01/24/2003 10:48:37
> > number of concurrent (pending )i/o might be increased, of course.
> > do you mean, it can pressure memory?
> 
> Ya.  I'm not sure it is niglectable or not.  Also, chance of pagermap
> shortage will increase (tho, this is not related to amount of physical
> memory).

i think 'struct buf' itsself is neglectable.
of course, it can be different with nfs, software raid or some complicate
things.  but IMO, if they're consuming much memory, they should concern
about memory shortage instead of genfs layer. (they can do sync i/o
even when genfs layer wants async i/o)

about pagermap, maybe we should throw it away and
do physical address based i/o. :-)

> Anyway, I guess we need to measure and improve if necessary.

ok, i'll do some benchmarks for nfs when i have enough time.
(hopefully this weekend)
i guess there're no noticable differences, though.

YAMAMOTO Takashi