Eric Bollengier wrote: > Hello, > > I've already think about this optimization, the only problem is that the > posix_fadvise argument isn't a bitwise flag. So you can't mix them like that.
Ok, I digged a bit more on google. I think the correct usage would be to call POSIX_FADV_SEQUENTIAL before reading the file and DONTNEED after having read it. (SEQUENTIAL should double the readahead window,DONTNEED will drop it from the pagecache). > Perhaps you can call this function multiple times, but the documentation is > unclear about this subject. Since NOREUSE will do something it'll also be "evil" to call before reading the file. This seems to be the best doc I've found. http://insights.oetiker.ch/linux/fadvise/ The operations gone through above fits quite well to how the file-daemon works. Then I found a link to discussing spool-file fadvises on bacula-devel over a year ago, but since most backup servers are dedicated this will only harm bacula itself. And I would expect the kernel to drop the spool files from memory when they subsequently get truncated anyway. http://www.mail-archive.com/[email protected]/msg03187.html -- Jesper ------------------------------------------------------------------------------ Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day trial. Simplify your report design, integration and deployment - and focus on what you do best, core application coding. Discover what's new with Crystal Reports now. http://p.sf.net/sfu/bobj-july _______________________________________________ Bacula-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/bacula-devel
