One way to know for sure what access method is being used is to get a GTF trace of at least one I/O request from your wayward job that you think should be counted. Turn on the IOSB option in GTF. You don't need to trace both SSCH and I/O interrupts; SSCH is probably enough. Find one trace record and look at the byte at IOSB driver ID byte. Sorry, but I don't remember its offset. I think it's 2. Look it up in the data areas to be sure. Check what IOS Driver did the I/O that was traced. If the IOSB driver id byte contains the value for an EXCP, then this I/O should and must have been presented to the SMF accounting routine. If it contains any other value, it will not automatically be presented to the SMF accounting routine as part of IOS processing, but the application program might have another way to account for it in SMF data.
Bill Fairchilde ----- Original Message ----- From: "Skip Robinson" <[email protected]> To: [email protected] Sent: Friday, December 13, 2013 11:12:39 PM Subject: Re: EXCP Counts in SMF Exit Aha. We may be on to something. The problem step in the problem job executes IBM 'high performance unload' utility PGM=INZUTILB. The step unloads/backs up a large DB2 data base. The job has almost two dozen steps, but we're focused on this one in particular. It's the only step in the job that has no device stats in job messages. . . JO.Skip Robinson Southern California Edison Company Electric Dragon Team Paddler SHARE MVS Program Co-Manager 626-302-7535 Office 323-715-0595 Mobile [email protected] From: Charles Mills <[email protected]> To: [email protected], Date: 12/13/2013 08:10 PM Subject: Re: EXCP Counts in SMF Exit Sent by: IBM Mainframe Discussion List <[email protected]> A triplet consists of a 32-bit offset into the record, a 16-bit length, and a 16-bit repetition count. For DB2 it gets *really* complicated, but for Type 30, it should be fairly simple. Perhaps your repetition count is more than one? If so, there will be another data section at the end of the first one, that is, at record+displacement+length. I read the doc as saying there may be more than one section, that is, SMF30EON might be non-zero and greater than one. You may need a loop on SMF30EON. Let me know if you want more detail. Charles -----Original Message----- From: IBM Mainframe Discussion List [mailto:[email protected]] On Behalf Of Skip Robinson Sent: Friday, December 13, 2013 3:37 PM To: [email protected] Subject: EXCP Counts in SMF Exit We use an SMF exit modeled on SAMPLIB(SCEACTMS). For each step, it formats various stats in the job message log for every device used. It works well most of the time, but for a particular job in a particular where we really need to see some values, the crucial lines are totally missing. I've looked at the code and concluded that the only way we could be missing all device detail is if the field SMF30EON is zero. The job step in question does enormous I/O to tape. PUTMSG3B DS 0H EXCP COUNTS SLR R5,R5 GET COUNT OF ICM R5,B'0011',SMF30EON EXCP SECTIONS <------------- BZ PUTMS3BX IF NONE, SKIP RTN <------------- BAL R14,MSGRTN PUT BLANK LINE OUT MVC MSGAREA,LINE3B SET UP HEADER LINE BAL R14,MSGRTN PUT LINE OUT BAL R14,MSGRTN PUT BLANK LINE OUT LR R4,R9 GET RECORD ADDRESS A R4,SMF30EOF POINT TO ID SEGMENT USING SMF30EXP,R4 ENABLE RECORD DSECT EXCPLOOP DS 0H PRINT NON-ZERO EXCP COUNTS PUSH USING USING LINE3BD,R6 ENABLE DETAIL LINE DSECT ... I've searched the web and found some references even in IBM-Main to a 'triplet' that handles multiple EXCP sections. But I haven't found any example of how to utilize the additional fields. I suspect that the data I need is available to the exit, but I need some guidance. ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
