Optimize oracle on VMware (April 2011)
Download as PPTX, PDF5 likes4,194 views
- Virtualizing Oracle databases on VMware ESX requires careful configuration and monitoring of memory, CPU, and I/O resources to avoid performance issues. - ESX uses memory overcommitment and techniques like ballooning and swapping to manage physical memory across VMs, which can cause problems if they page Oracle memory like the SGA. - Multiple virtual CPUs on a VM must wait to be scheduled on the physical CPUs, showing up as ready time and impacting performance of CPU-intensive workloads. - Dedicating storage, using raw device mapping (RDM), and tuning I/O can help optimize storage performance for virtualized Oracle databases.
Optimize oracle on VMware (April 2011)
- 1. Optimize Oracle RDBMS on VMwareGuy HarrisonDirector, R&D Melbournewww.guyharrison.netGuy.harrison@quest.com@guyharrison
- 5. AgendaMotivations for VirtualizationVMware ESX resource management:MemoryCPUIOParavirtualization (OVM) vs Hardware Assisted Virtualization (ESX)
- 7. Resistance to Database virtualization
- 8. DB virtualization is happening
- 10. Oracle virtualization is lagging....
- 12. Managing ESX memoryESX can “overcommit” memorySum of all VM physical memory allocations > actual ESX physical memoryMemory is critical to Oracle server performanceSGA memory to reduce datafile IOPGA memory to reduce sort & hash IO ESX uses four methods to share memory:Memory Page Sharing“Ballooning”ESX swappingMemory compressionDBA needs to carefully configure to avoid disaster
- 13. Configuring VM memoryVMs Compete for memory in this rangeRelative Memory Priority for this VMMaximum memory for the VM (dynamic)Minimum Memory for this VM
- 15. ESX and VM memoryESX SwapESX swapVMESX virtual memory Effective VM physical memory ESX physical memoryVM virtual memory
- 16. ESX BallooningESX SwapESX swapVmmemctl“balloon”VMESX virtual memory Apparent VM physical memory Effective VM physical memory ESX physical memoryVM SwapVM Swap
- 17. ESX BallooningAs memory grows, ESX balloon driver (vmmemctl) forces VM to page out memory to VM swapfile
- 18. ESX BallooningInside the VM, paging to the swapfile is observed.The guest OS will determine which pages are paged outIf LOCK_SGA=TRUE, then the SGA should not be paged.
- 19. ESX SwappingESX SwapESX swapESX virtual memory VMEffective VM physical memory ESX physical memoryVM virtual memory
- 20. ESX SwappingESX SwapESX swapVMApparent VM physical memory Effective VM physical memory ESX virtual memory ESX physical memory
- 21. ESX SwappingESX swaps out VM memory to ESXswapfile
- 22. ESX SwappingWithin the VM, swapping cannot be detected.ESX will determine which memory pages go to diskParticularly occurs when VMware tools are not installedEven if LOCK_SGA=TRUE, SGA memory might be on disk
- 23. Avoiding Ballooning and swappingmemory reservations help avoid ballooning or ESX swapping
- 24. Other VMware memory management thoughtsMemory page sharing Multiple VMs can share an identical page of memory (Oracle code pages, etc) Modern chipsets reduce memory management overheadMultiple hardware page layersMemory compression (new in ESX 4.1)Pages are compressed and written to cache rather than to diskSwapping is more expensive than ballooningSlower to restore memoryOS and Oracle get no choice about what gets paged“Double paging” can occur – guest and ESX both page a block of memory
- 25. Ballooning vs. Swapping Swingbench workload running on Oracle database – from VMWare whitepaper: http://www.vmware.com/files/pdf/perf-vsphere-memory_management.pdf
- 26. VMware memory recommendations Paging or swapping of PGA or SGA is almost always a Very Bad Thingtm.Use memory reservations to avoid swappingInstall VMware tools to minimize swappingBallooning >> swapping (double paging, paging SGA, high swap-in latency, etc)Set Memory reservation = PGA+SGA+process OverheadBe realistic about memory requirements:In physical machines, we are used to using all available memoryIn VM, use only the memory you need, freeing up memory for other VMsOracle advisories (or Spotlight) can show you how much memory is neededReduce VM reservation and Oracle memory targets in tandem to release memory
- 29. VMware CPU management If more virtual CPUs than ESX CPUs, then VCPUs must sometimes wait.Time “stops” inside the VM when this occurs For multi-CPU VMs, it’s (nearly) all or nothing). A VCPU can be in one of three states:Associated with an ESX CPU but idleAssociated with an ESX CPU and executing instructionsWaiting for ESX CPU to become available Shares and reservations determine which VM wins access to the ESX CPUs
- 30. Configuring VM CPU VMs compete for CPU in this rangeShares determine relative CPU allocated when competing
- 31. CPU utilization VM“CPU Ready” is the amount of time VM spends waiting on ESX for CPU Inside the VM, CPU stats can be misleading
- 32. SMP for vCPUsESX usually has to schedule all vCPUs for a VM simultaneouslyThe more CPUs the harder this isSome CPU is also needed for ESXMore is therefore not always better(Thanks to Carl Bradshaw for letting me reprint this diagram from his Oracle on VMWare whitepaper)
- 33. ESX CPU performance comparisonsVT enabledVs 2 core 1.8 GHz physical machine
- 36. ESX CPU recommendations Use up to date chipsets and ESX softwareAllocate as few VCPUs as possible to each VMUse reservations and shares to prioritise access to ESX CPUPerformance of CPU critical workloads may be disappointing on older hardwareMonitor ESX Ready time to determine the “penalty” of competing with other virtual machines
- 37. ESX IO managment
- 38. Typical VMWare disk configuration
- 39. Disk Resource AllocationDisk shares can be used to prioritize IO bandwidth.However, ESX often does not know underlying storage architecture
- 40. Performant VMware disk configuration
- 41. Optimal configuration See “Oracle Database Scalability in VMware® ESX” at www.vmware.com/oracleEach virtual disk directly mapped via RDM to dedicated RAID 0 (+1) group 41 Spindles!
- 42. ESX IO recomendationsFollow normal best practice for physical disksAvoid sharing disk workloads Dedicated datastores using VMFS Align virtual disks to physical disks?Consider Raw Device Mapping (RDM)If you can’t optimize IO, avoid IO:Tune, tune, tune SQLPrefer indexed pathsMemory configurationDon’t forget about temp IO (sorts, hash joins)
- 43. Shameless plugs
- 47. Paravirtualizationvs “Hardware Virtualization”Virtualization is not emulation....Where-ever possible, Hypervisor runs native code from OS against underlying hardwareBecause a virtualized operating system is running outside privileged x86 “ring 0”, direct calls to hardware need special handling.The three main approaches are:Full Virtualization (VMWare on older hardware)ParaVirtualization (Xen, Oracle VM)Hardware Assisted Virtualization (Intel VT, AMD-V)
- 48. Full virtualization Hardware calls from the Guest are handled by the hypervisor by:Catching the calls as they occur at run timeRe-writing the VM image at load time (binary translation)Requires no special hardwareSupports any guest OSRelatively Poor performanceUsed by ESX on older chip-setsGuest OSHypervisorRing 0Hardware
- 49. Hardware Assisted virtualizationIntel VT and AMD-V chips add a “root mode”.Guest can issue instructions from non-root Ring 0.CPU can divert these to hypervisorNo changes to OS requiredGood performanceRequires modern chipsetsRoot ModeNon-Root ModeGuestHypervisorRing 0Hardware
- 50. ParavirtualizationGuest operating system is rewritten to translate device calls to “hypercalls”
- 51. Hypercalls are handled by a special VM (dom0 in Xen/OVM)
- 52. Good performance but requires modified guest OS
- 53. Xen can use either paravirtualization or hardware assistGuest(domU)Ring 0Guest(dom0)HypervisorHardware
- 54. RAC and ESX
- 55. Paravirtualization, ESX and RACPrior to 11.2.0.2, Oracle relied on paravirtualized kernels to maintain time synchronization for RAC clusters.From 11.2.0.2 Oracle uses Cluster Time Synchronization Service (CTSS) to maintain clock sync, and this works on ESXTherefore, Oracle supports RAC on Vmware ESX only from 11.2.0.2 onwardsSee Oracle MySupport Note 249212.1
- 56. ReferencesLatest version of this presentation:http://www.slideshare.net/gharriso/optimize-oracle-on-vmware-5271530My blog (www.guyharrison.net ):http://guyharrison.squarespace.com/blog/2010/2/22/memory-management-for-oracle-databases-on-vmware-esx.htmlhttp://guyharrison.squarespace.com/blog/2010/4/9/esx-cpu-optimization-for-oracle-databases.htmlhttp://guyharrison.squarespace.com/blog/2010/7/12/stolen-cpu-on-xen-based-virtual-machines.htmlhttp://www.vmware.com/files/pdf/perf-vsphere-memory_management.pdfhttp://www.vmware.com/files/pdf/Oracle_Databases_on_vSphere_Deployment_Tips.pdf
Editor's Notes
- #3: Apologies, I’m a database type.....Quest is best known for toad, but we also have enterprise monitoring across all levels of the stackIn Melbourne, SQL Navigator + the spotlights. It’s not a complete co-incidence about the star trek theme.
- #4: Had a lot of fun with toad over the years
- #5: Insanely popular – literally millions of users
- #14: As well as total memory to the VM, you can:Adjust memory “shares”, which determine priority for this machine when in contention with other machinesReservation: guaranteed (sort of) amount of memory to allocate Limit, if you want to prevent VM from getting all it’s memory Memory reservationsMemory SharesIdle memory taxMemory sharingThe Balloon driver (vmmemctl) ESX swapping