Phil Pennock, 2011-10-31 08:41: > You've mounted the filesystem with "atime" support, which reports last > access time. While useful, for busy file-systems this default support > in Unix has proven to be a historical mistake (in my opinion). > > The only reason the disk is being affected here is because the act of > reading the file is updating the inode with a new atime, and this needs > to be written back to the disk. Otherwise, assuming local disk and not > NFS, the file would sit in buffer cache and all new reads would never go > to disk, because the cache would still be valid.
I second your opinion. atime is kind of a broken concept. OTOH, atime updates are cached (AFAIK), so the impact should only be visible on very busy systems with little RAM. > Mount the filesystem noatime or move the file to a filesystem which is > mounted noatime and see how that affects performance. btw, Linux uses relatime by default since 2.6.30. > CDB is probably the way to go then. CDB (and DBM etc.) cannot be used with iplsearch. We would need a trie data structure for that. Don't know if there are widely used standard formats and tools for that... -- ## List details at https://lists.exim.org/mailman/listinfo/exim-users ## Exim details at http://www.exim.org/ ## Please use the Wiki with this list - http://wiki.exim.org/