vmware clusters basics, pros, cons, possible radicl implementation by: nathan krussel
TRANSCRIPT
VMWare Clusters
Basics, Pros, Cons, Possible RADICL implementation
By: Nathan Krussel
Basics(Based on ESXi 5.0 & Vsphere 5.0)
• A VMWare cluster is composed of at least 2 ESX severs– There is a maximum of 32 hosts per
cluster
• vCenter has the ability to manage multiple clusters–Maximum of 1000 hosts per vCenter
Sever– 10,000 max powered on VM’s• 30,000 VM’s across 10 linked vCenter
instances
Basics (cont)
• Per Host (ESX) maximums–Max of 160 Logical Processors• 25 virtual CPU’s per core• 2048 Virtual CPU’s per host
– 2 TB of RAM– Physical NICS (2 to 32)
Basics (cont)
• Cluster Maximums– 32 max hosts (DRS enabled or not)• 8 max if there are over 40 VM’s per host• 16 max if HA and DRS are both enabled
– 3000 VM’s per cluster• 512 per host
–Maximum Failover rate• 100%
Basics (cont)
• HA clusters– High Availability– Allow for failover with “Instant recovery”– Keeps cluster up and running with out
losing machines until replacement server arrives
–When failover occurs machine is cold booter
Basics (cont)
• DRS clusters– Allow for live migration (through Vmotion)– Automatic load balancing*– Keeps for “overextending” a host– No recovery based on failed server
Basics (cont)
• HA/DRS clusters– High availability and reliability• If one host fails, VM’s are migrated accordingly
– Has survivability of HA and load balancing of DRS to make a stable and responsive cluster.
– Sharable resource pool– Requires no additional resources over non
cluster based VI (Virtual Infrastructure)*
Pros (Based on HA/DRS Clusters)
• Allows for high availability, always running even in the event of a server failure– Can add/remove/replace host with out
powering down the entire cluster
• Load balancing between all hosts and clusters*–Most users won’t notice the transition
between hosts, very short pause
• No manual moving of VM’s neccesary
Pros (cont)
• Expandable infastructure• Easily upgrade host hardware with out
reconfiguring cluster• Alerts you to failures in the cluster• Automatically will shift VM’s if you put
a host into maintenance mode
Cons(Based on HA/DRS Clusters)
• High initial investment cost• Under utilized machines eat resources• More noise• Requires more A/C• May not know exactly which host has
which VM’s
RADICL Implementation
• This is a possible design for what could be done to RADICL
• Changes can be made based on funding levels and time alotted
RADICL Implementation
• Hosts– 12 GB+ of RAM per host– 2 CPU’s with at least 4 cores per socket– Proper cooling to prevent over heating– 4 Ethernet ports per host
• Cluster– 4 to 8 hosts (or more depending on need)
RADICL Implementation
• Miscellaneous – 2 Data stores accessible by entire cluster• Fast RAID array (1+0) preferably solid state
drives for active machines• Templates on a slower raid array
– Fast vCenter machine to handle all the creation of the VM’s
– Additional program for power management of servers (optional)
References
• http://www.vmware.com/pdf/vsphere5/r50/vsphere-50-configuration-maximums.pdf
• http://www.petri.co.il/vmware-esx-server-cluster-pools.htm• http://www.vmware.com/products/vcenter-server/features.html