On Fri, 21 May 2010 13:41:25 -0500, Al Cole wrote: >In all of the FTP data processes that I create, I require that the users edits >and maintains the data file in their application directories. >When they have completed creating the file, the user or the application will >move the file to an FTP server. By doing so, the user or application is >stating that the file is complete and ready for processing. > An excellent idea.
My bias is toward a UNIX server. Where can Ed get a UNIX server? z/OS is UNIX! Why UNIX? So that Ed can copy the file to a temporary pathname then rename() it to the target name. POSIX rename() has some properties remarkably suited to this purpose: o rename() is preemptive. Any preexisting file with the target name is quietly and unconditionally replaced. o rename() is atomic. POSIX requires it to be so serialized that no other process can perceive an instant when the file does not appear to exist. o rename() (and unlink()) obey a sort of LUW isolation. If the FTP server begins reading the "old" copy and the rename occurs while the transfer is in progress, a reference count guarantees that data blocks will not be discarded/overwritten until the FTP server issues a close() for the descriptor of that file. (PDSE works somewhat similarly.) Guaranteed integrity with minimal serialization protocol impacting the user. -- gil ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
