At 15:42 13-12-02, McKown, John wrote: >I know that I should "use the code, Luke", but I'm not that familiar with >the kernel et al. Does Linux/390 take advantage of any of the VM facilities >when running under VM vs. in an LPAR? I'm thinking especially of the >"handshaking" that is possible with paging. I.e. Linux thinks the page is in >memory, but VM has it paged out. I think this is done with VSE and I >remember it back in the OS/VS1 days as well. What about other VM-only >facilities?
Yes, pseudo page fault support is there already. When a process gets blocked because the particular page is paged out by VM, the kernel gets a chance to run another process. Recent changes to z/VM improved the PFAULT support. I have not seen numbers about how effective this is for Linux, but it is enabled by default when you run in a virtual machine. In fact, virtual memory itself as provided by z/VM is already a benefit over LPAR since it allows the Linux guests to breathe. The other thing is the shared kernel support that has been there for some time now. This allows you to put some 2MB of the kernel in shared pages and thus reduce the footprint of your penguins. And the dasd driver can use Diagnose I/O instead of SSCH and exploit MDC and other z/VM benefits. There have been some problems with that part of the driver in the past, but it looks like the current code works. Rob
