Addressing Ted's reply: No, there is no startup overhead. The STC is already running before I start the test, and still running when I end the test. The "start and end processing" is exactly the same for every record, whether the previous one was 1 or 100 ms ago. Well, the program comes back to a WAIT ECBLIST after every record, so I guess there is a little less overhead if one of the ECBs is already posted.
Addressing Elardus's reply: Believe it or not, there is an amazing amount of program there. It turns out mainframes are really, really fast! <g> There is a tiny assembler IEFU8x exit that makes a quick decision "do we want this record or not?" I'm not counting the CPU time for that code as it runs on the TCB of the SMF record producer, or the SMF task, or on an SRB -- not sure, but in any event, it's not part of the STC. The STC that processes an SMF record in .000083 seconds is written in C++ (other than that WAIT ECBLIST and some associated code) and does a heck of a lot of work. It does a whole lot of record formatting with code that is all table-driven, not classic "hard" field coding, and produces 300 to 2000 bytes of formatted message. It calls Unicode services to translate the message to UTF-8 and then ships it out via TCP/IP. All in .000083 seconds (z196)! Charles -----Original Message----- From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf Of Ted MacNEIL Sent: Friday, January 31, 2014 10:19 AM To: [email protected] Subject: Re: CPU time There is probably a start-up overhead that is fixed, regardless of the transaction rate. So, the more transactions running the lower this per transaction cost appears. Example: Assumptions: Fixed overhead: 10 seconds Cost/transaction: 1 second For one transaction: 10+1 = 11 seconds/tx For two: 10+2 = 12 divide by two: 6/tx ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
