On 10/26/2016 07:05 AM, Paul Flint wrote: > In '86 land the OS, say Linux, essentially eats all the memory, and > makes it virtual (as opposed to virtuous :^). ...
Linux does the same on z: it just parks on all memory it sees. > In zVM land you have a kick ass memory manager that essentially lies > to each Virtual Machine and tells it that the memory limit is in the > Exabytes (gee, I love that word :^). The guest operating system on > the Virtual Machine in turn uses this lie to set the limit the docker > engine can operate a docker instance based upon the lie it got from zVM. z/VM is a hypervisor. All hypervisors "lie" to their guests about how much memory is installed. You see the same thing with VMware, KVM, VirtualBox, Hyper-V, ... any of them. > Kinda makes you wonder how hard it would be to make the docker engine > talk directly to zVM and get the lie directly from the lier... Trivial. And conceptually the same for any hypervisor because they all "lie" about memory for the sake of establishing a fake "machine" which the guest sees. (z/VM provides a direct interface which the bare metal doesn't have, but the guest could use either that or the bare metal method to figger it out.) > If I had a zVM account somewhere or an emulator that legally ran the > current zVM I could look into this, right now I just want to know if I > am following the your issue correctly. I likely need to find out more > about docker clusters, but right now I have a hunch the answer to > better clustering is on the zVM side. I should probably reply on the other thread. I /think/ what Phil is trying to do is get some measurements from Docker on z for the sake of sizing "production" hosting with Docker. So if I am understanding, Phil needs to incrementally scale up the memory seen by this one specific z/VM guest. But adding memory on-the-fly to a guest usually requires a reboot of the guest. So others on the list have suggested a couple of the usual tricks for tweaking memory w/o rebooting the guest. Docker is a good candidate for running Linux "heavy". That is, Docker could use Linux in its own LPAR and without z/VM. (KVM is another such candidate.) Even so, LPARs get lied to by PR/SM about how much memory is present. (Seems these lies abound!) So Phil will want to continue his measurements to know how big the Docker LPAR(s) should be. -- 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://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/