On Thu, Jan 13, 2005 at 12:29:01AM +0200, guy keren wrote: > in fact, it's very hard to achieve a proper "which process caused this > I/O" log. consider the case where two processes wrote to the same position > in a file - there's likely to be only one disk write operation - which of > the two processes will you account this I/O against?
Whichever caused this block to be read into the page cache. > also, what about read-ahead? when a process reads data from the disk, the > operating system typically performs a read-ahead. in fact, sometimes the > mere opening of a file will cause the VFS layer to perform read-ahead of > data for the file - even if your application didn't read anything. That's IO caused on behalf of the process. > thus, you should try to _properly_ define what is it that you're trying to > account for, including those 'multiple updates - single disk I/O' > and 'read ahead' cases. I fully support this statement :-) Cheers, Muli -- Muli Ben-Yehuda http://www.mulix.org | http://mulix.livejournal.com/
signature.asc
Description: Digital signature