VSP 3067 Mythbusters Goes Virtual Name Title Company
VSP 3067 Mythbusters Goes Virtual Name, Title, Company
Mythbusters Goes Virtual Mattias Sundling Evangelist, Quest Software Eric Sloof VMware Certified Instructor, NTPRO. NL
Introduction • v. Sphere evolves with every release. • Things that used to be true aren't true anymore. • Engage in virtualization communities and social media to get up to speed.
Agenda/Myths 1) i. SCSI RDMs have superior performance over VMFS. 2) CBT causes significant overhead on your VMs. 3) Resource Pools should always be used to categorize and allocate resources to VMs. 4) LSI SCSI is always better than Paravirtual SCSI.
Myth 1: i. SCSI RDMs have superior performance over VMFS
What is a RDM RAW LUN VMFS volume • • • A RDM is a symbolic link. When a LUN is opened for access, the mapping file is read to obtain the reference to the raw LUN. Thereafter, reads and writes go directly to the raw LUN rather than going through the mapping file.
VMkernel and Storage • Physical Compatibility mode RDM • Virtual mode RDM • Regular VMDK on VMFS • Logical device IO scheduler • v. Storage APIs • Adapter IO scheduler and Linux emulation
The i. SCSI Laboratory • Iomega Stor. Center px 6 -300 d with 6 SATA 7200 Disks • Windows 2008 R 2 4096 MB – 1 v. CPU Hardware Version 8 • VMware v. Sphere 5 • Single Intel 1 GB Ethernet • Cisco 2960 switch MTU Size 1500
Results Passmark Performance. Test • • RDM Physical RDM Virtual VMDK on VMFS Microsoftware ISCSI initiator
Myth Busted • 1 Gigabit/sec (Gbps) = 112. 16 MB/s. • Use VMFS datastores whenever possible. • VMFS is optimized for storing and accessing large files. Only Use RDMs if your VM • is performing SAN snapshotting. • is clustered to a physical machine using MSCS. • has large amounts of data (64 TB) that you do not want to convert into a VMDK (2 TB).
Myth 2: CBT causes significant overhead on your VMs
What is CBT? • Driver inside VMkernel • Identifies change blocks within virtual disks • Block size based on VMDK size • Backup window and host resources significantly reduced • Requirements: v. Sphere 4+ and Virtual HW v 7+ • Limitations: p. RDM, i. SCSI within VM • Enable through v. Center or backup application CBT Driver
CBT overhead • Memory – Max 256 KB / Disk (2 TB) – 0. 125 KB / GB VMDK • CPU – Turning on a bit in bitmap when an I/O request completes • Storage – Space • . ctk file 0, 5 MB / 10 GB VMDK – I/O • Every time disk gets closed, change tracking info written to disk
Test/Result • Very little overhead CPU, Mem, Storage • Could not measure it, except. ctk file • No negative impact on disk I/O CBT Disabled CBT Enabled
Myth Busted • CBT overhead is very small • Backup window and host resources reduced significant • Always use CBT if your backup solution supports it
Myth 3: Resource Pools should always be used to categorize and allocate resources to VMs
Resource Settings • CPU – Shares, Reservation, Limit. • Memory – Shares, Reservation, Limit. v. App Resource Pool VM
Sibling CPU Shares Value • v. App = 4000 • Resource Pool = 4000 • VM = number of v. CPU * 1000
Sibling Memory Shares Value • v. App = 163840 • Resource Pool = 163840 • VM = Mem MB * 10
The Laboratory VM 4000 Shares 6 Ghz RP 4000 Shares 6 Ghz v. App 4000 Shares 6 Ghz 0, 75 Ghz 0. 5 Ghz 1. 5 Ghz Root Resource Pool 24 Ghz 6 Ghz
Myth Busted If you want to • use Resource Pools, do your math! • group VMs in a container use the blue folders! • have apple (VM) and orange (v. App) siblings take care of them!
Myth 4: LSI Logic SCSI is always better than PVSCSI
What is PVSCSI? • Paravirtual SCSI – Introduced in v. Sphere 4, improved in v. Sphere 4. 1 – Designed for high performance (+12%) – Requires less resources on v. Sphere Host (-18%) – Supports only Win 2003+, RHEL 5+, SUSE Linux Enterprise 11 SP 1+, Ubuntu 10. 04 +, Linux 2. 6. 33+ – Virtual HW v 7+
Test Result LSI Logic vs. PVSCSI Logic PVSCSI • Very similar disk throughput • Lower Host CPU Utilization for PVSCSI Host CPU utilization 90 MB/s – 60 000 IOPS
Myth Busted PVSCSI is equal or faster than LSI SCSI PVSCSI requires less Host resources Used to have more limitations PVSCSI is better in all ways so why aren´t we using it on all supported VMs? • Takes time to change behaviour • •
Summary • • VMFS is better than RDMs CBT has very little overhead Do your math when using sibling RPs PVSCSI is equal or faster and requires less resources than LSI SCSI BUT…
References • CBT – https: //www. vmware. com/support/developer/vddk/vadp_vsphere_backup 12. pdf • PVSCSI – http: //kb. vmware. com/selfservice/microsites/search. do? language=en_US&cmd=display. KC&external Id=1017652 – http: //kb. vmware. com/selfservice/microsites/search. do? language=en_US&cmd=display. KC&external Id=1010398 – http: //blogs. vmware. com/performance/2009/05/350000 -io-operations-per-second-one-vspherehost-with-30 -efds. html – http: //www. vmware. com/pdf/vsp_4_pvscsi_perf. pdf – http: //www. vmware. com/pdf/Perf_Best_Practices_v. Sphere 4. 1. pdf
Questions ? Mattias Sundling Evangelist, Quest Software mattias. sundling@quest. com, @msundling Eric Sloof VMware Certified Instructor, NTPRO. NL esloof@ntpro. nl, @esloof
- Slides: 28