> >What I, personally, think would be "neat" would be some sort of R/O "page > >data set". It would be something like the LPA. It would be created and > >updated to basically contain the z/OS nucleus. NIP would simply allocate > >the fixed real memory for this "nucleus". It would read the data into this > >memory from this "page data set" . It would then create the page tables > >necessary to properly map the real memory into shared virtual memory. The > >"page data set" would actually be maintained by some z/OS "system > >generation" program from the contents of SYS1.NUCLEUS, plus "anything else" > >that NIP uses to create the z/OS nucleus. NIP would also read in the HCD > >infomation for the UCBs et al. > > > >But, again, this will cost up-front money to design and implement. What we > >have now works; works rather well; and is reliable. > > > >What the above is based upon is vaguely based on the VM concepts of NSS or > >DCSS segments. > > > IPL in milliseconds! I notice you cautiously suggest to "allocate > the fixed real > memory..." up front, avoiding rude storage management surprises. Perhaps > good to verify the integrity of the image early, also. > > And no need for PDSE in NIP, nor for PDS ever.
With modern processors and DASD, the time spent loading the nucleus is such a small part of system initialization time that it is not worth spending significant development resources to make it faster. How small is it on your system? Under IPCS, do IPLDATA STATUS Jim Mulder z/OS System Test IBM Corp. Poughkeepsie, NY ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
