Messages in this thread | | | Date | Sun, 21 Jan 2001 23:20:13 -0200 (BRST) | From | Marcelo Tosatti <> | Subject | Re: mmap()/VM problem in 2.4.0 |
| |
On Fri, 19 Jan 2001, Rik van Riel wrote:
> On Thu, 18 Jan 2001, Marcelo Tosatti wrote: > > On Thu, 18 Jan 2001, Rik van Riel wrote: > > > On Fri, 12 Jan 2001, Vlad Bolkhovitine wrote: > > > > > > > You can see, mmap() read performance dropped significantly as > > > > well as read() one raised. Plus, "interactivity" of 2.4.0 system > > > > was much worse during mmap'ed test, than using read() > > > > (everything was quite smooth here). 2.4.0-test7 was badly > > > > interactive in both cases. > > > > > > Could have to do with page_launder() ... I'm working on > > > streaming mmap() performance here and have been working > > > on this for a week now (amongst other things). > > > > Also remember that drop_behind() is not working for mmap(), yet... > > filemap_sync(..., MS_INVALIDATE) needs a 2-line change to have > drop-behind. I have this running (more or less) on my laptop here.
The filemap_sync() modifications on your patch makes msync (MS_SYNC|MS_INVALIDATE) not work correctly anymore. filemap_sync() must set the page dirty for all valid pte's in the requested area if MS_SYNC is set.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
| |