- What one needs to consider when AFD runs on a NFS filesystem

     Remember the internal directory structure of AFD and the
     what happens internally when it sends a file. Make sure that
     the files directory (and maybe also the archive directory) is
     also located on the NFS mount.

     When you monitor directories on an NFS filesystem and you have
     problems that AFD does not pick up files, use the [dir options]
     'force reread'. NFS sometimes does not update the date of the
     directory correctly. 'force reread' will force AFD to read the
     content of the directory every time during each scan. Otherwise
     it would just check the date of the directory to see if anything
     changes.

     For performance and stability reasons it is not good if fifodir
     is on an NFS mount. Better to use tmpfs or ram disk for the fifodir.
     This has the disadvantage that you will loose some statistical
     information after a system crash, but you will not loose any data.


   - Things to consider when configuring the FTP server

     A good and secure FTP server is vsftpd. Do not run the FTP server
     via xinetd, it is very slow. Also, disable logging if you do not
     need it. Logging is very expansive on a system that receives lots
     of files.

     Use only FTP accounts that have no shell. The password is transmitted
     in clear text via internet or RMDCN!

     chroot the FTP account so that the FTP user cannot explore your
     server.


   - Let AFD only move files and NOT copy them

     As mentioned earlier always let AFD run as closely as possible
     where the actual data is produced. Ensure that at least that the
     files directory is in the same filesystem where it's input
     directories are. Use soft links.
   
   - Some performance tweaks

     Compile in only that what you need.

     Use the configure option --enable-compiler-optimizations

     On linux (as of kernel 2.6.17) use --enable-splice_support.
     This avoids copying data from kernel space to user space
     when copying files locally. The copying only happens in kernel
     space.

     For network connections use --enable-sendfile_support. Again the
     kernel does not need to copy any data to user space for any sockets.
     Only tested on linux.

     Do NOT do the following:

        $AFD_WORK_DIR/etc/DIR_CONFIG

        [directory]
        ~afd/data/in

           [files]
           *

              [destination]

                 [recipient]
                 file://afd@local/data/out1
                 file://afd@local/data/out2
                 sftp://afd@local/data/out3
                 ftp://ftponly:secret@local/data/out4

                 [options]
                 exec -d bzip2 -9 %s

     Rather do it as follows:

        $AFD_WORK_DIR/etc/DIR_CONFIG

        [directory]
        ~afd/data/in

           [files]
           *.bz2

              [destination]

                 [recipient]
                 file://afd@local/data/in

                 [options]
                 exec -d bzip2 -9 %s
                 lock OFF

           [files]
           !*.bz2
           *

              [destination]

                 [recipient]
                 file://afd@local/data/out1
                 file://afd@local/data/out2
                 sftp://afd@local/data/out3
                 ftp://ftponly:secret@local/data/out4

     Mount your filesystem with the noatime flag! Most newer linux
     systems have relatime set automatically. But better is noatime.

   
   - What one needs to consider in an High availability environment

     First thing you need to decide if you really need a cluster with
     a shared filesystem/disk. Arguments against it are higher complexity
     and single point of failure. If you decide that you need it, have
     a look at DRBD (http://www.drbd.org/).

     For cluster software have a look at heartbeat (http://www.linux-ha.org).
     Note, that you do not need pacemaker (keeps your applications running).

     If you use heartbeat or pacemaker there is a script in the AFD
     source package that does all the necessary checks and starts/stops
     AFD during a switch over. The script can be found in the directory
     script and is also just called afd. Since this script is also
     used for starting AFD when system boots, you have to rename this
     to afdha to activate it's HA features. In the case of heartbeat
     you would just need to issue the following command:

        cp scripts/afd /etc/ha.d/resource.d/afdha