System z ntroduction to ization SHE naheim, Session 14561 omney White, BM romneyw@us.ibm.com System z rchitecture and echnology 2013 BM Corporation
genda ntroduction to ization Concept Server ization pproaches Hypervisor mplementation Methods Why ization Matters ization on System z artitions Machines 2 System z ization echnology 2013 BM Corporation
ization Concept esources roxies for real resources: same interfaces/functions, different attributes May be part of a physical resource or multiple physical resources ization Creates virtual resources and "maps" them to real resources rimarily accomplished with software or firmware esources Components with architecturally-defined interfaces/functions May be centralized or distributed - usually physical Examples: memory, disk drives, networks, servers Separates presentation of resources to users from actual resources ggregates pools of resources for allocation to users as virtual resources 3 System z ization echnology 2013 BM Corporation
Server ization pproaches Hardware artitioning Bare-metal Hypervisor Hosted Hypervisor pps S... pps S pps S... pps S pps S... pps S artition Controller djustable partitions Hypervisor Hypervisor Host S SM Server SM Server SM Server Server is subdivided into fractions each of which can run an S partitioning S/370 S-to- and -to-s Sun Domains, H nartitions partitioning BM eserver pseries L H vartitions Hypervisor provides fine-grained timesharing of all resources Hypervisor software/firmware runs directly on server System z L and z/vm WE Hypervisor VMware ESX Server Xen Hypervisor Hypervisor uses S services to do timesharing of all resources Hypervisor software runs on a host operating system VMware GSX Microsoft Server H ntegrity VM KVM Characteristics: Bare-metal hypervisors offer high efficiency and availability Hosted hypervisors are useful for clients where host S integration is important Hardware partitioning is less flexible than hypervisor-based solutions 4 System z ization echnology 2013 BM Corporation
Hypervisor mplementation Methods Virt Mach L S rivp L... rap and Emulate rap VM runs in user mode ll privileged instructions cause traps Hypervisor rivp emulation code ranslate, rap, and Emulate Virt Mach L S rapp L... rap VM runs in user mode Some -32 instructions must be replaced with trap ops Hypervisor rivp emulation code Examples: Benefits: ssues: C-67, VM/370 uns unmodified S Substantial overhead Examples: Benefits: ssues: VMware, Microsoft VS uns unmodified, translated S Substantial overhead Hypervisor Calls ( aravirtualization ) Virt Mach L S Hcall L... Call VM runs in normal modes S in VM calls hypervisor to access real resources Hypervisor service Direct Hardware ization Virt Mach L S rivp L... Exit VM runs in normal modes Hardware does most of the virtualization (SE architecture) Hypervisor provides control Hypervisor service Examples: Benefits: ssues: WE Hypervisor, Xen High efficiency S must be modified to issue Hcalls Examples: Benefits: ssues: System z L, z/vm, Xen High efficiency, runs unmodified S equires underlying hardware support 5 System z ization echnology 2013 BM Corporation
Server ization Business Value oles: Consolidations Dynamic provisioning / hosting W orkload management W orkload isolation Software release migration Mixed production and test Mixed S types/releases econfigurable clusters Low-cost backup servers Servers ization Server ossible Benefits: High resource utilization Great usage flexibility Enhanced workload QoS High availability / security Low cost of availability Low management costs Enhanced interoperability Legacy compatibility nvestment protection ization can fill many roles and provide many benefits n the final analysis, its potential benefits take three forms: Help reduce hardware costs Help increase physical resource utilization Small footprints Can improve flexibility and responsiveness resources can be adjusted dynamically to meet new or changing needs and to optimize service level achievement ization is a key enabler of on demand operating environments such as cloud Can reduce management costs Fewer physical servers to manage Many common management tasks become much easier 6 System z ization echnology 2013 BM Corporation
System z ization est z/vm Web Sphere HiperSockets & etworking and Switching CMS Control Memory FL s L Sysdmin ools rogram pache / & etwork WebSphere WLM z/s Memory L s raditional L and Batch WLM z/s Memory L esource / System Manager (/SM) F z/vm est z/s Memory L ntelligent esource Director (D) Multi-dimensional virtualization technology System z provides logical (L) and software (z/vm) partitioning /SM enables highly scalable virtual server hosting for L and z/vm virtual machine environments D coordinates allocation of CU and / resources among z/s and non-z/s Ls* * Excluding non-shared resources like ntegrated Facility for processors 7 System z ization echnology 2013 BM Corporation
L CU Dispatching and Execution Control CU rogram nstruction stream SE: Start nterpretive instruction Execution nstructions roblem state nstructions High-Frequency Control nstructions that require virtualization Low-Frequency Control nstructions that require hypervisor virtualization SE nterception to hypervisor E.g., Modify Subchannel Load, Store, dd,... Start Subchannel, est Subchannel,... L hypervisor SE Hardware nstruction Execution Controls L CU SE Descriptor Hardware or Firmware nstruction nterpretation Handling ization ssists Load Store dd SSCH SCH CU nstruction Execution Unit 8 System z ization echnology 2013 BM Corporation
L Dispatching z/s artition z z z/vm artition z z z/vm artition FL FL z/s & DB2 z/s & DB2 JV & XML DB2 z z FL FL L hypervisor dynamically dispatches: 1. -purpose logical processors on general-purpose physical processors 2. z logical processors on z physical processors 3. z logical processors on z physical processors 4. FL logical processors to FL physical processors z z z z FL FL - ool z ool z ool FL ool 9 System z ization echnology 2013 BM Corporation
L Memory artitioning collection of up to 2 31 - or 2 64 -bytes virtual address spaces Each virtual machine may have its own virtual address spaces mapped to this common logical partition address space L 1 L 2 L z/vm Machine Memory Space z/vm Hypervisor z/s or High- erformance artition Memory Spaces artition Hypervisor = the real partition memory pages associated with a virtual address space; that is, the sets of dynamically-allocated physical memory pages necessary to run a z/s task or a process collection of up to 2 64 -bytes virtual address spaces z/s and exploit multiple such virtual address spaces = the real partition memory pages associated with a virtual machine; that is, the sets of dynamically-allocated physical memory pages necessary to run a guest operating system in a virtual machine 10 System z ization echnology 2013 BM Corporation
L Multiple Channel Subsystems CU Channel Subsystem - mage 1 Channel Subsystem mage 2 Channel Subsystem mage 3 Channel Subsystem mage 4 System z Channel Subsystem FC Switch, CU - Devices, etc FC Channel ath ransparently by ll artitions S Ethernet dapter ransparently By artitions Configured to Channel Subsystem mage Switches, etwork Links, etc. 11 System z ization echnology 2013 BM Corporation
L High-erformance / Sharing artition 1 artition 2 artition 3 artition / dapter Subchannel mage For artition 1 Subchannel mage For artition 1 Subchannel mage for artition 2 Subchannel mage for artition 3 Subchannel mage for artition Channel ath mage 1 Channel ath mage 2 Channel ath mage 3 Channel ath mage dapter nterface (e.g., a FC channel path) he / operations for each logical partition are multiplexed within the adapter/channel path and on the associated / interface Storage Controller Device Device Device Device Device Device Volume e.g., a arallel ccess Volume (V) he / infrastructure (adapters/channels, their transmission links, and attached / resources are shared by Ls at native speeds (without hypervisor involvement) / requests, their associated data transfers, and / interruptions flow between each S instance and the shared / components, just as if the / components were physically dedicated to a single S instance 12 System z ization echnology 2013 BM Corporation
z/vm CU Dispatching and Execution Control CU rogram nstruction stream SE: Start nterpretive instruction Execution nstructions roblem state nstructions High-Frequency Control nstructions that require virtualization Control nstructions that require hypervisor virtualization SE nterception to hypervisor E.g., Start Subchannel Load, Store, dd,... Set Storage Key, Signal,... z/vm hypervisor SE Hardware nstruction Execution Controls z/vm CU SE Descriptor Hardware or Firmware nstruction nterpretation Handling ization ssists Load Store dd SSKE SG CU => CU nstruction Execution Unit 13 System z ization echnology 2013 BM Corporation
z/vm Dispatching z/s Guest z z z/vm Guest z z Guest FL FL z/s & DB2 z/s & DB2 JV & XML DB2 z z rocess rocess z/vm hypervisor dynamically dispatches: 1. -purpose processors on general-purpose processors 2. z processors on z or general-purpose processors 3. z processors on z or general-purpose processors 4. FL processors to FL or general-purpose processors z z z z FL FL - ool z ool z ool FL ool 14 System z ization echnology 2013 BM Corporation
z/vm Memory ization L 1 Machine 1 Machine 2 Machine z/vm Machine Memory Space z/s or Machine Memory Spaces z/vm Hypervisor z/vm Hypervisor artition Hypervisor = the guest real memory pages associated with a virtual address space; that is, the sets of dynamically-allocated host virtual memory pages necessary to run a z/s task or a process = the guest real memory pages associated with a virtual machine; that is, the sets of dynamically-allocated host virtual memory pages necessary to run a guest operating system in a virtual machine 15 System z ization echnology 2013 BM Corporation
z/vm Disk ization 1 /W Minidisk Cache (High-speed, in-memory disk cache) /W Disk in Memory /W Disk in Memory /W / / B /W 1 pplication 2 Code 3 /W C z/vm DSK: dynamic temporary disk allocation pool DSK 1 Minidisk DSK space Fullpack Minidisk B Minidisk C ESS 750, ESS 800, DS8000, DS6000 Minidisk: real disk partitioning technology Legend: /W = ead/write / = ead-nly 16 System z ization echnology 2013 BM Corporation
z/vm L and Switch ization Guest C Guest C Guest C Guest C Guest C Guest C VM Switch Controller ort 65 ort 66 ort 67 ort 68 ort 69 ort 70 Load Balancer ggregator / Multiplexer LC ort 1 ort 2 ort 3 ort 4 z/vm Switch z/vm System z L S S S S LC ort 1 ort 2 ort 3 ort 4 (Link ggregation Control rotocol) eal Switch 17 System z ization echnology 2013 BM Corporation
z/vm perations utomation 1. Send all console output to a single CMS virtual machine server farm 2. Use and EXX to interrogate console messages EXX erformance oolkit for VM ealtime Graphs CMS Console Console Console Console CMS eports, Historical Data 3. ssue hypervisor commands on behalf of servers C Hypervisor perations C Monitor 1. Use the C Monitor to automatically capture performance and resource consumption data for each server Monitor Data 2. Use erformance oolkit for VM to process Monitor data 18 System z ization echnology 2013 BM Corporation
System z ization Genetics ver 45 years of continuous innovation in virtualization efined to support modern business requirements Exploiting hardware technology for economical growth C-67 S/360 VM/370 S/370 Control rogram Hypervisor Dynamic ddress ranslation (D) 308x 303x 4381 VM/S Diagnose Hypervisor nterface SM Zone elocation VM/H 64 MB eal zseries VM/ES 9672 VM/X 31-Bit EXX nterpreter 1960s 1972 1980 1981 1988 1995 2004 z/vm BM System z a comprehensive and sophisticated suite of virtualization function ES 64-Bit Machine esource Manager Disks in Storage CMS ipelines ccounting Facility bsolute elative SHE Discontiguous Saved Segments nstruction CE rogrammable perator () nter-user Communication Vehicle (UCV) Conversational Monitor System (CMS) 9x21 3090 rogram Event ecording (E) ranslation Look-side Buffer (LB) Start nterpretive Execution (SE) L Hypervisor Dedicated / s VM ssist Microcode System z9 Guest Ls Minidisk Cache SE on SE amed Saved Systems Business Value: Scalability, eliability, obustness, Flexibility,... Expanded Storage Multiple mage Facility (MF) Switch Set bserver erformance oolkit / riority Queuing dapter nterruption ass-hrough Multiple Channel Subsystems (LCSS) pen Systems dapter (S) etwork Switching Large SM System z10 QD Enhanced Buffer State Mgmt ntegrated Facility for z196 HiperSockets zec12 utomated Shutdown Host age-management ssist HyperSwap Dynamic Machine imeout _ort D ization (V) Live Guest elocation SS SM xc penstack 2013 19 System z ization echnology 2013 BM Corporation