On Wednesday, 09/18/2002 at 02:12 AST, Adam Thornton <[EMAIL PROTECTED]> wrote: > Wed, Sep 18, 2002 at 01:50:05PM -0400, Alan Altmark wrote: > (Steve Shultz wrote) > > > 2) Your storage should be defined to whatever size you want as the minimum > > size of the virtual machine to ipl your shared system. I believe it would > > be unwise to define it smaller than 64M. > > Why is this? I've seen VM tell me I didn't have enough memory for a > shared CMS segment as well. This question is really because I'm a VM > ignoramus, but why do shared segments require more memory? Is it simply > that they're mapped into your machine in largish chunks and therefore > you lose some R/W storage with every R/O shared segment you define?
When you are populating the shared segment, rather than just using it, your virtual storage size MUST be large enough to "cover" the segment (segments are sized in 1MB increments). And, yes, you need to allow that nMB of R/W storage will be lost. Also, I don't think Linux yet understands discontinguous chunks of guest real storage. That is, it would be upset if there were "holes" in the middle of storage as would exist if there were a gap between the end of R/W storage and, say, the start of the R/O kernel. (CMS, on the other hand, was designed with NSSs and DCSSs in mind.) > > When its done, you'll be in CP READ and you or anyone else will be able to > > issue "CP IPL system_name" to start up your shared system. One caveat. > > Every virtual machine that ipls your shared system must have the same disk > > configuration as the system that was saved. That is, the disks must be at > > the same addresses and be the same sizes as the virtual machine that > > issued SAVELX2. > > Hmmm. Is this simply a matter of the savesys including a parmfile > within it? In which case, could you get around this by, say, defining a > small root volume, and then dynamically bringing DASD online in the > boot-scripts? Yes, you could do that. The NSS could include a R/W or R/O ramdisk if you wanted. But Steve's point is that the content of storage must be the same. You can manipulate starting conditions any way you want, just be sure they are identical for all users using the NSS. (I think the saved starting point is in entry.S.) I encourage people to experiment with this support and to come to a consensus as to how it can best be used. This was just a first step in getting Linux for zSeries to use z/VM's proven shared-memory model. Alan Altmark Sr. Software Engineer IBM z/VM Development
