The big difference between "System Z" (and all things akin to that architecture, regardless to re-branding, I can't keep up!) and the other platforms is that Z has _channelized I/O_. I've always suspected, but never heard or seen proven or disproved, that channelized I/O makes virtualization potentially more effective/efficient on Z. I do know that the "insertion loss" of virtualization is way lower with Z than any other.
I was not aware of true virtualization for some of the architectures you cite (SPARC, MIPS). I heard that Sun offered something akin to PR/SM, but never heard of a /hypervisor/ for SPARC. And today, it's all about containers, which are nothing more or less than 'chroot' on steroids. There's really only ever >one< kernel in a container host. But containers (as a service, feature, or concept) are available for /any/ HW architecture where Linux runs (and a few others, such as FreeBSD, see "jails"). Linux itself supports more instruction sets / architectures than I can count. One Linux release I happen to have on-hand lists more than two dozen: i386, x86, x86_64, s390, arm, powerpc, but also alpha, cris, ia64, m68k, mips, parisc, score, sh, sparc, unicore32, xtensa. Some of these have dropped off (this is an old source), and most I've honestly never seen (hardware). But you're asking about true virtualization, where one can run a /different/ kernel/nucleus in the "guest" (which one CANNOT do in a container). I'm aware of two true "type 1" hypervisors: z/VM and Xen. Originally, Xen was actually PARAvirt, but when the newer instructions became available on AMD/Intel, Xen kinda grew up. As I recall, Xen had to boot before Linux. I ran Xen at home with good success, but didn't delve into the details. At the time, I believe Xen was limited to Intel/AMD. There are more "type 2" hypervisors than I can count. Perhaps the most popular these days is KVM on Linux, which supports _Z_, _AMD/Intel_, _ARM_, and _POWER_. Haven't used Qubes. My wife's nephew outed that he's using Parallels on his Mac. Apple has used POWER, now Intel/AMD, and soon ARM, so Parallels (a type 2 hypervisor) is in the same space as KVM (is it type 2 or type 1?) w/r/t supported HW instruction sets. I thought RISC-V was "close enough" to POWER that it might also run KVM. Does this help? -- R; <>< On 10/30/21 5:34 AM, CAREY SCHUG wrote: > Hi Bill from the distant past.... > > I am NOT NOT NOT asking you to take you time to explain to me, just point to > existing documents (including any you may have written that you can share). > warning: I have been out of mainframes for 20 years and had no formal > training on x86 computers. > > My ONLY question to you is the goal below. Are you aware of any printed > books, downloadable documents, other list servers where I can ask, or can you > suggest how I could craft a web search for: > > Wanted-- > > An explanation (and comparison) of x86 vs IBM virtualization, for a person > with ONLY zVM background. > > Including glossary of terms (like what we called core cancer, t think they > call memory leak). > > Hopefully explaining x86 ring levels beyond their existence level that I am > aware of. > > Optional bonus: A comparison x86 vs ARM, and within x86, AMD vs Intel. And > are there add-on hardware memory managers that might not easily be identified > when I walk into a computer store to buy one? As to which is "better", or do > each have advantages in some areas? I hadn't thought about this till I > started this email, and have found some promising articles, but so far all > written for somebody whose vocabulary base is x86...which is like middle > english would be to me. I have not written in machine code for small > computers beyond the Z-80. > > Also I would like details on malware exposures and how to protect the > hypervisor from them. For instance, it seems to me that "buffer overrun" > (though historically mostly winblows) could in theory happen in any intel > based system since unlike mainframes, the hardware does not hard block the > end of the input buffer. > > I read about X86 type 1 vs type 2 hypervisors, but then details of some > purported type 1 sound more like type 2 to me. Then I found Qubes (and I > think parents, children and siblings of it) which at first glance sounds like > the most extreme type 1 possible given the x86 memory architecture. But it > seems Qubes is still not complete. And maybe to be secure, to I need to have > multiple Ethernet adapters, one for each guest? Or maybe running a linux > firewall in a read-only guest would suffice? > > personal disappointment: Wikipedia seems totally ignorant of any > virtualization other than IBM-z/x86/sparc/arm/power, while every other > mainframe manufacturer I presume has some form, HP, Digital, I think I even > heard that some big cisco routers virtualization, and other IBM product > lines, but maybe they were other processors under the covers, such as later > AS/400s being power processors. Some quick searches show many of them > migrating to ARM, MIPS, etc, so maybe not. Except there was a reference to > MIPS virtualization, which is not in the table in wikipedia. Oh, MIPS is > dead, maybe RISC-V? Quick search seems to indicate there is no working > hypervisor for RISC-V yet, but it is in development? > > --Carey > > ---------------------------------------------------------------------- > For LINUX-390 subscribe / signoff / archive access instructions, > send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit > http://www2.marist.edu/htbin/wlvindex?LINUX-390 -- -- R; <>< ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit http://www2.marist.edu/htbin/wlvindex?LINUX-390
