On Fri, 5 Sep 2008, Marcelo Leal wrote: > 4 - The last one... ;-) For the FSB allocation, how the zfs knows > the file size, for know if the file is smaller than the FSB? > Something related to the txg? When the write goes to the disk, the > zfs knows (some way) if that write is a whole file or a piece of it?
For synchronous writes (file opened with O_DSYNC option), ZFS must write the data based on what it has been provided in the write so at any point in time, the quality of the result (amount of data in tail block) depends on application requests. However, if the application continues to extend the file via synchronous writes, existing data in the sub-sized "tail" block will be re-written to a new location (due to ZFS COW) with the extra data added. This means that the filesystem block size is more important for synchronous writes, and particularly if there is insufficient RAM to cache the already written block. For asynchronous writes, ZFS will buffer writes in RAM for up to five seconds before actually writing it. This buffering allows ZFS to make better informed decisions about how to write the data so that the data is written to full blocks as contiguously as possible. If the application writes asynchronously, but then issues an fsync() call, then any cached data will be committed to disk at that time. It can be seen that for asynchronous writes, the quality of the written data layout is somewhat dependent on how much RAM the system has available and how fast the data is written. With more RAM, there can be more useful write caching (up to five seconds) and ZFS can make better decisions when it writes the data so that the data in a file can be written optimally, even with the pressure of multi-user writes. Bob ====================================== Bob Friesenhahn [EMAIL PROTECTED], http://www.simplesystems.org/users/bfriesen/ GraphicsMagick Maintainer, http://www.GraphicsMagick.org/ _______________________________________________ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
