This page includes technical information about different aspects of the virtualization service:
The Server Virtualization Services utilizes state-of-the-art hardware from HP. A typical host server will have (4) quad-core processors for a total of 16 cores per host. Coupled with 128GB of RAM on each host, our infrastructure can support many virtualized guests per host. The hardware is clustered in an N+1 setting where we have enough computing power to have at least 1 ESX host fail and still handle the load to run all of the VM guests. All servers have dual power supplies which are fed from different circuits and also include UPS and generator backup power. The local disks utilize RAID technology so losing a single disk will not cause downtime. The hosts are also monitored 24x7 by our management systems and will notify staff if there are any errors.
We use VMware's Enterprise ESX product as the basis for our Server Virtualization Service. ESX allows us to create a cluster of host servers that give us our needed N+1 level of redundancy. We also utilize Distributed Resource Scheduling (DRS) to allow the VM guests to automatically migrate between the ESX hosts based on our predefined limits of memory and processor performance. For example, if a particular ESX host is "busier" than other hosts, DRS can automatically migrate (vmotion) a VM guest to a less busy ESX host. This gives the guest VMs the best and most balanced performance. We also utilize the High Availability (HA) option for ESX. In the event that a running host server has a catastrophic failure, all of the guest VMs that were running on that host will automatically restart on one of the other hosts in the cluster thereby greatly reducing the downtime from a catastrophic failure. Each VM guest is also protected by an OS snapshot that allows us to recover a VM to a previous version of the operating system if needed.
Each of our ESX servers is connected in a redundant configuration to at least 2 network switches through 5 server class network ports on the server. We could lose either a network switch or a network card and still deliver the necessary network throughout required. We are currently running our network connections at 1 Gbps with the ability to move to 10Gbps when needed.
We have different tiers of storage for various types of VM guests we host. The main storage pool is a fully redundant Network Attached Storage (NAS) which uses fiber-channel disks. We have the ability to grow the underlying storage as needed as well as the ability to resize a guest VM disk. The NAS utilizes a redundant copy of the data between our 2 datacenters so we could have an entire NAS fail and still be able to quickly make the changes to bring up the VM guests on the redundant NAS. For less IO intensive disk needs, we also have iSCSI based storage available. With either storage option, we have the ability to offer shadow copies, mirror copies and tape backups of the data as needed.
We offer access to the virtual center (VC) console (via our campus-wide virtual desktop service) to allow named administrators the ability to connect to the virtual console of their guest VM and perform power management functions (Power On/Off). Specific guest performance data is also available via the VC console. By using ESX, we can also take a snapshot of the guest VM on an as-needed basis. This allows the customer to take a snapshot prior to a software upgrade or other major changes to a guest VM. After the upgrade is completed, we have a choice to accept the changes or to roll back to the state of the VM at the time of the snapshot. This allows a safe rollback option in the event that a planned upgrade does not work as desired.