On Sun, Oct 17, 2010 at 5:44 PM, William H. Blair <[email protected]>wrote:
> Sam Siegel explains: > > <--snip--> > > I located a copy of (the very old) SA22-7619-1 and tried to find what you > might be referring to. > My bad on the document revisions number. It should be -18 (SA22-7619-18) for v1r10 <--snip--> > > There is not one single reference to either IWM4CON or IWM4ECRE. There is a > reference to IWMCLSFY. But the string "zIIP" does not appear in this > document in any place whatsoever (not that Adobe Reader can find, that is). > Could you point out the _exact_ page where you think this information is? > > Sorry about that ... it was a typo on my part as mentioned above. I did not intentionally mean to waste your time. > Regardless, on the assumption that the "-1" was a typo on your part, I > looked at my copy of SA22-7619-20 for z/OS 1.12. There is nothing in there > about zIIPs (in this context), either. > Agreed ... zIIPs are not mentioned specifically in that context. However, it is what is not said that gave me an idea to experiment on. > I again looked at IBM's licensed documentation on how one does this. Here's > all I can say: You're not even close. Yes, you do have to use an SRB and > yes, you do have to use an enclave. But doing just those two things is not > sufficient to get an SRB dispatched on a zIIP. You have to do more. And > that > "more" is the subject of the IBM proprietary, licensed material, which you > do not have (not even the documentation). > Without the documentation, I cannot make any statements with respect to this. > > > In looking at SMF30_ENCLAVE_TIME_ON_zIIP field in the SMF30CAS > > section, I'm seeing a non-zero value for the jobstep. > > I'm clueless. Put an SRB that you think is running on a zIIP into a long > (enabled) loop, first on a regular CP (to measure how long it actually > takes > to complete your loop), and then -- you think -- on a zIIP. See if your > enclave time on zIIP value goes up to what you would expect it to be. > > But to really prove that your SRB is running on a zIIP, cause a system dump > to occur inside your SRB routine and examine the dump to see which > processor > it was actually executing on. Compare that to the "+I" processor ID > exhibited by the D M=CPU operator command. > I will do the second option and report back as that provides the most conclusive evidence. > > It is extremely unlikely that you have accidentally stumbled upon the > mechanism to make this little bit of magic happen. It is always possible > that there is a bug in z/OS and your SRB is, in fact, being dispatched on a > zIIP. But I seriously doubt that. > Agreed, it could be a bug. But I've seen this in v1r9 and v1r10. I'm not saying that there is certainty that the SRB is dispatched to a zIIP all of the time. But it seems to happen extremely regularly. > -- > WB >
