Workstation is 70 - 90 percent of native performance

GSX is 70 - 90 percent of native performance

ESX is 83 - 98 percent of native performance

WKS, ACE, GSX - Maximum device compatibility

ESX - Maximum performance, lowest overhead, dynamic resource control, virtual snmp

GSX + ESX - up to 3.6GB of ram per VM

ESX - up to 80 VCPU's per ESX, dynamic resource allocaion, vmotion, uses VMFS

VirtualCenter - centrally manage all vm's, deploy using standard templates, monitor state and performance. Manage access to virtual machines using one point of control.

VMotion - move running vm's from one server to another with no downtime.

VirtualCenter - runs only on windows

VMWare management interface or mui is web based

Remote console runs on windows or linux

VMKernel manages all running virtual machines

Service Console supports server side of client interfaces. e.g mui, remote console and VirtualCenter

Some features have only command line interface

P2V - takes a snapshot of exisiting physical system. Transforms to a VM - does not support linux.

VMWare benefits - Server consolidation, disaster recovery, testing and development, desktop deployment

vmkpcidivy -i to reassign nic to the console OS

Service console needs 192mb of ram for 8 vm's, 172mb ram for 16 vm's, 384mb for 32 vm's or 512mb for more than 32 vm's.

Can only have 4 primary partitions, if you want more than 4 you need to make three primary and the fourth extended

Configuration file for mapping the service console file system to mount points is /etc/fstab

service console swap should be twice the amount of memory assigned to the service console

/boot lives on the main boot disk, 50mb, ext3, used by the service console and boot

(non) on main boot, 2gb ext3 used by the service console and swap

/home on main boot, 2gb ext3, used by the service console and homes

/ on main boot, 2gb ext3, used by the service console and root

/vmimages can be on any disk, size varies, ext3, used by the service console, templates and iso images.

(none) on any local, 100mb, vmkcore, used by vmkernel and core dump

(automatic) on any local, depends on the amount of ram overcommitment, vmfs-2, used by vmkernel and swap

consider making /tmp and /var seperate partitions

VMFS allows multiple access - more than one server can access simultaneously

VMFS used for VMDK (virtual disks), VSWP (vmkernel swap), VMSS (VM Suspend Files) and VMTD (VirtualCenter templates)

VMFS has near raw speed for virtual disks, safe for large files, and safe simultaneous access by many esx hosts.

vmkernel swap allows overcommitment of ram, and can be activated in the mui after creation

vmfs public means many esx can use the volume but only one esx can use the file at the same time

vmfs shared means many esx can use the same file at the same time but is only used for clustering

virtualcenter uses the VPXuser account on esx server

virtualcenter default port to talk to esx is 902

each vm has 6 pci slots - 1 is used for video card. scsi and ethernet cannot be greater than 5

Disk Modes

• Persistent - has no redo log
• undoable mode provider prompt at power off and power on about what to do with the redo log
• append mode accumulates without asking
• Redo logs slow down performance

Network Drivers

• VLance maps to universally supported nic
• VMXNet offers high performance while creating less cpu load but driver only comes with VMware tools

Remote console uses SSL over TCP on port 902

VMware tools installs drivers for video card, mouse, highest speed networking, optimised bus logic scsi and memory management.

VMware tools also installs heartbeat service, manages scripts to run in guest OS, time synch between guest and service console and shrink to zero all unallocated space in virtual disk.

San storage and ESX allows ease of moving vm's, clustered applications, disaster recovery, and template repository

VMKernel cannot use virtual disks on NAS (does not understand NFS or SMB)

NAS can be used for /vmimages storage

VMKernel can use virtual disks on SAN (SAN required for clustering)

VMKernel can use raw luns on SAN (required for physical to virtual clustering)

Allocate HBA's to VMKernel to make SAN available to VM's, then scan for new luns

Disk.Maxlun governs how many luns to scan. If lun n to be scanned set Disk.Maxlun to n + 1

Disk.Masklun enables lun ranges to be ignored on boot. e.g Disk.Masklun = vmhba0:0:1-14, 17-31

Multipathing allows continued access to SAN lun in the event of a hardware failure

Failover Policies

• MRU - Most recently used will continue to use the last active path
• Fixed - or preffered path will revert back when available

VMFS Volumes are flat file systems, no subdirectories and are limited to 256 files per volume

VMFS lives under /vmfs

vmkfstools -x 8192 vmhba0:0:6:1:mydisk.vmdk will grow mydisk.vmdk to 8192mb

ESX uses SCSI reservations to lock luns

vmfs has metadata and metadata gets updated by

• creating files
• changing file attributes
• power on
• power off
• growing virtual machines

metadata update process locks vmfs for duration of process

each vm can have up to 4 virtual nics

esx can have up to 9 gig or 16 10/100 nics

vmware nic ranges : 00:0C:29:*.*.* or 00:50:56:*.*.*

virtual switch with one outbound adapter is a vmnic

32 vm's can connect to one vmnic and there are zero collisions on internal traffic which can be controlled by traffic shaping

/etc/vmware/netmap.conf documents vmnic mapping to pci address

/etc/vmware/devnames.conf documents network labels of each virtual switch

Virtual switches with > than 2 outbound adapters are bonds

Bonds connect to 802.3ad nic team

up to 32 vm's per bond, zero collisions on internal traffic, improved performance by network load balancing and redundant nic operation.

bandwidth in bonds can be controlled by traffic shaping

up to 10 physical nics on a virtual switch can be teamed in a bond

/etc/vmware/hwconfig documents the physical nic that makes up a bond, see below for specific

• OUT-MAC is the default each vm's nic is mapped to the physical nic based on the vm's mac. It has a low overhead is compatible with all switches and may not spread the traffic evenly
• OUT-IP is a nic for each outbound packet is chosen based on the source and destination IP. This enables better distribution of traffic, has a higher cpu overhead and not compatible with all switches
• STANDBY uses one home nic exclusively and will fail over to another if needed

Virtual switches with no outbound adapters are "vmnets"

vmnets provide networking for vm's of a single esx host only, up to 32 vm's and traffic shaping is not supported.

Resource optimisation

• CPU - Minimum rate, maximum rate, processor affinity and share allocation
• Disk - Share allocation
• Nic - Traffic Shaping

Single vcpu runs on 1 physical cpu at a time

dual vcpu runs on 2 or 0 cpu's

transperant page sharing allows pages with identical contents to be stored only once

Maximum absolute cpu percentage 200 percent for dual cpu vm's. 100 percent for single cpu vm's

Processor affinity to LCPU if hyperthreading. PCPU if not hyperthreading

vmkfstools -e /vmimages/b.vmdk (destination) local:a.vmdk (source) -> correct syntax!

• -e makes template
• -i reverse source and destination and will deploy from a template

monolithic file is just one BIG file

COW file is a file broken into segments e.g 10 x 2gig files

vmotion requirements

• source and destination must have visibility to all luns used by the vm, gig ethernet backplane, access to the same physical network, consistently labeled virtual switches, and identical cpu's
• source and destination must NOT have connections to internal switches or vmnets, must not be clustered, must not have connected removeable devices  and open remote console sessions.

vmkusage - a web based graphical display of performance data every 5 minutes

esxtop - runs at the service console command prompt and gives live performance view updated every 5 seconds

"ready time" was time when the vm was ready to do work but no cpu was available

host based alarm : send email, send notification trap and run script

VM based alarm : send email, send trap, run script, suspend vm, power off vm or reset vm

Tools for monitoring the service console

• uptime - tabulate cpu usage
• top - reveal cpu intensive processes
• vmstat - reveal paging activity
• vdf -h will reveal free disk space
• free - tabulate service console ram and swap
• find - find new and large files
• man - display command descriptions and syntax

vmsnap.pl is a snapshotting backup script written in perl

Clustering

• cluster in a box - consists of multiple vm's on a single esx host, protects against operator error, application crash and operating system crash
• cluster across boxes - cluster vm across esx hosts, shared storage is required, protects against operator error, application crash, operating system crash and hardware failure
• Cluster between physical server and VM - ESX is standby host, low cost n+1 redundancy, shared storage is required.
• Clustering ESX servers requires veritas cluster server

Best Practices

• Test memory for 72 hours before deployment
• Install a dummy OS on your hardware to test
• Check supported hardware list
• Allocate enough resources to the service console
• Develop datacenter policies

Troubleshooting

• A shortage of service console ram will cause : poor remote console performance and poor mui performance
• A shortage of service console swap will cause : randomly killed processes (especially mui) and cannot start new vm's
• A shortage of service console disk will cause : inability to launch the MUI, and incomplete logs.
• If you cannot start a vm check if there is enough vmkernel memory, is there enough vmkernel swap, sufficient priviledges, permissions on the virtual disk, is the root file system full or is there no swap space
• If a vm hangs or bluscreens is there a guest os or app problem, bad ram, or vm software bugs.
• If an application is performing poorly check the environmental issues and if the key app resources are limited.