Close Window



Know more about this button?

OK, the problem began when Charlie Whitman and I were waiting in the lobby of a hotel which was the locus of an interim SHARE, about 15 months before the button appeared. The hotel was in the southeast, if you have a list of old SHARE venues. Charlie mentioned that many 3083 users of SAS (where he worked had very poor VM simulation of priv ops resulting in high T/V ratios (total/virtual CPU time) and I would look more at the problem when I got home. I said that 3083s were famous for having very high T/V ratios. I sat down at a terminal and set SPOOL PRINT START TO * and TRACE PRIV and ran the SAS installation test then closed the virtual printer. A PRINT and I had a spool file of several hundred thousand lines that appeared in virtual reader. In a minute or so, I had found the location of an SSK (set storage key) instruction that was responsible for almost all the privs in the trace. I no-oped the instruction in my nucleus and reran the test. The T/V ratio was quite normal (low) and the test completed much more quickly. I called Charlie with the info, and we concluded that the key management was the problem. In a couple of hours he called me back and said that CMS did a lot of key management on GETMAIN and FREEMAIN macro simulation. We discussed it and concluded it was unnecessary. Charlie wrote a source update (Barry: BECAUSE BACK THEN WE HAD 100% SOURCE CODE OF BOTH VM AND MVS, BEFORE OCO) to eliminate the SSK instruction from the CMS nucleus, and we both started testing it. No ill effects were found, so we opened a performance APAR with IBM, assigned number VM14999. It was shortly closed WAD (Working As Designed). Meanwhile my installation (which did not even have a 308x processor) and Charlie's and many dozens of others started running the deletion code as a mod to their systems. When I challenged IBM about their closure of the problem, they said that CMS was designed that way, even though the simulation did NOT match the behavior of real hardware. When I asked if every installation was running the mod, would IBM then incorporate it into the product, the IBMer assigned to the CMS Project said NEVER. It then became an us vs. them issue, and was a major plank in the anti-OCO movement. HOPE THIS HELPS. DAVE GOMBERG, 2011.
1 Items Returned