Vmware Basics

Front 1

3 considerations when planning a vmware install

Back 1

1. how many virtual machines
2. what resource will they use
3. what mix of applications

Front 2

Licensing is recommended to be set at the time of the install though it can be changed later. What are the two types of licenses?

Back 2

1. server-based is the default configuration, licenses stored in a single file
2. host-based allows licenses to be stored in files located on each ESX system

Front 3

vSwitch

Back 3

-each vm communicates with each other and the outside through the vSwithc
-has embedded security
-multiple connections form a NIC team. NIC teams distribute traffic and provide passive failover
-a vm can be multihomed by connecting to more than one vSwitch. These can be used as routers and firewalls.

Front 4

templates or golden masters

Back 4

are standard virtual machines used for cloning. Templates cannot be powered on.
Templates should contain all the tools and applications necessary. (Guest os, VMware Tools, Mgt agents, Anti-virus, Backup agents, Applications, etc.)
Used to quickly build virtual machines.

Front 5

Setup to optimize performance

Back 5

1. validate hardware (test memory, run benchmarks, check compatibility matrix)
2. allocate the right amount of resources (memory and disk space)
3. use supported guests
4. install the latest version of
VMware Tools (vmkfstools)
5. install VMware's BusLogic driver (the VMware driver has optimizations that the guest os does not)
6. disable screen savers, animations, xServer (on Linux)
7. disconnect unused devices on the guest and host (floppy, usb, lpt, com, etc.)
8. schedule backups and anti-virus for off-peak hours

Front 6

CPU Performance Best Practices

Back 6

1. Allow for CPU overhead, avoid excessively overcommitment of process
2. use as few virtual CPUs (VCPUs) as possible
3. multi-threaded/process applications in an SMP guest, it helps to pin guest threads/processes to VCPUs

Front 7

Memory Performance Best Practices

Back 7

2 kinds of ESX overhead: 1. additional time to access memory within a vm 2. extra memory for ESX server code and data structures

2 components of memory overhead 1. fixed system-wide oh for the service console, VMkernel, and for each vm 2. the number of virtual CPUs, the guest os's configured memory and 32 or 64-bit guest os

1. VMkernel reclaims memory by ballooning and swapping. (avoid frequent reclamation)
2. Correctly configure virtual memory to begin with
3. if possible use less than 896MB on Linux guest
4. ESX Server can still be overcommitted but not swapping (swap reduces perf)
5. is swap cannot be avoided place swap on high speed/bandwidth storage system

Front 8

Networking Performance Best Practices

Back 8

1. understand your nw: nw hops between systems, NICs, configurations (autonegotiate, speed, duplex), packet size (are packets fragmented or re-assembled)
2. multiple nw adapters, single vSwitch to a phy new form a NIC team.
3. 32-bit guests in an AMD PCnet32 defaults to VMware vlance driver (e1000 for 64-bit) but vmxnet provides better perf.
4. use separate vSwitches to avoid contention between the COS, VMkernel, and vms.
5. VMkernel nw and swtiches must be of the same config.
6. NIC morphing replaces vlance with vmxnet for improved perf

Front 9

Storage Performance Best Practices

Back 9

1. storage perf issues are most often config issues with the storage devices, not ESX
2. Major factors: workload, hw, vendor, RAID levels, cache sizes, stripe size, etc.
3. use esxtop to check i/o queue in the vmkernel 'QUED'
4. spread i/o loads over the available paths (across multiple HBAs and SPa)
5. for Active/Active arrays with fixed failover policy, designate perferred paths to each logical unit of storage.
6. avoid excessive open/close operations on VMFS (watch tail file_on_vmfs_partition)
7. Protect root from becoming full

Front 10

Virtual Disk Modes

Back 10

1. Independent Persistent - immediate changes and permanently written to disk
2. Independent Non-Persistent - changes discarded at reboot
3. Snapshot - captures entire state of vm (memory and disk) - allows you to revert back to that point in time.

Front 11

Storage Performance Best Practices cont'd

Back 11

8. load balance
9. fibre channel san outperforms NAS and iSCSI due to fibre channel protocol
10. avoid excessive file locks or meta data locks (caused by dynamically growing vmdk files and file perm changes)
11. configure max queue depth for HBA cards
12. increase vm's max outstanding disk requests (if needed)
13. avoid oversubscribed ethernet links (causes dropped nw packets)
14. applications that write a lot of data should have multiple connections to storage (avoid nw)

Front 12

ESX Disk Formats

Back 12

1. Thick - all space allocated at creation time
2. Eager zeroed - all space allocated and zeroed at creation
3. Lazy zeroed - zeroes out the block on first write
4. Thin - space required for thin-provisioned is allocated and zeroed on demand

Front 13

Resource Managment Best Practices

Back 13

1. If you expect frequent changes to total avai resources use Shares not Reservations.
2. reservations are the min amount of cpu or memory needed not the amt to have available
3.to fully isolate a resource pool, make the type: fixed, use reservation and limit.
4.group vm's for a multi-tier service in a resource pool
5. hyperthreading allows a single phy processor to behave like two logical processors

Front 14

3 Memory Management Mechanisms

Back 14

to expand or contract the amount of memory dynamically allocated. ESX Server host uses a proprietary transparent page sharing technique to securely eliminate redundant copies of memory pages (preferred method)
1. page sharing - similar os's load the same pages into memory, a single copy is used in memory and shared
2. ballooning - (preferred over swapping) vmmemctl module in the guest os reclaims pages that are considered the least valuable by the guest os
3. swapping - forcibly reclaims memory when paging and ballooning fail to reclaim memory. Configure memory reservations

Front 15

DRS Best Practices

Back 15

1. keep vm's together to improve latency but separate the vm's if needed for failover (if both machines have the same functions)
2. homogeneous systems perform better in clusters
3.