On Tue, Jan 22, 2002 at 09:57:10PM -0800, Swapnil Nagle wrote: > Greg KH <greg@kroah.com> wrote : > > > What are you going to use to predict what files to prefetch? Is this > > for an embedded system, where you know your workload, or for a "general" > > kernel improvement? > > We are using automatic predictive prefetching. A kernel thread will > collect the file open & read statistics and will generate a > probability graph. This graph is used to select the files to prefetch. I've seen code that _knew_ what files every executable was going to access, and still performance was not improved (the files were specified in a configuration file.) For 2.0.x kernels yes, but for 2.2.x and greater, no. That code is available somewhere in the SubDomain kernel patch on the www.immunix.org web site if you are interested. Good luck, greg k-h -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ IRC Channel: irc.openprojects.net / #kernelnewbies Web Page: http://www.kernelnewbies.org/
