Protection with High Availability

In many cases virtual servers are implemented as part of a server consolidation strategy. As a result creating a highly available environment is critical to the survival of any organization, no one can afford to be down with so many applications dependent on the server.

Virtual environments are susceptible to server failure with loss of access to applications and data. For example, each time a virtual server is moved to a different machine--such would be the case in a disaster recovery scenario--the connection would have to be interrupted and reconnected to the new hardware. VMotion is a fantastic tool to ensure that the applications move transparently, but what about the datastore?

Storage Virtual Appliances can be mirrored, presenting a single datastore that manages two copies of the data—one on each ESX server—that allows seamless failover. This is achieved using synchronous mirroring.

Storage Virtual Appliances running on different virtual servers mirror their datastores. Any datastore can be mirrored between SVAs running on different physical servers. ESX sees a mirrored volume as a single datastore with multiple access paths, an “Active-Active” model. The mirrored datastore can be accessed down any path. Writes, SCSI reservations etc are mirrored across to other mirror plex(es). If a mirror plex fails, ESX simply uses another path to the datastore. With this architecture and VMotion in place, the virtual environment can withstand a server failure or storage failure without disruption to the user.

  • Simple SVA is not enough
  • Virtual Environment is vulnerable to failure of server running SVA
  • SVA must be able to withstand the failure of a VMware server or its storage
  • This is achieved using synchronous mirroring
  • SVA running on different virtual servers mirror datastores
  • Any storage volume can be mirrored between SVAs running on different physical servers

High Availability though Synchronous Mirroring:

  • Volumes are mirrored by SVA
  • VMware sees a mirrored volume as a single virtual disk with multiple access paths
    • Active-active model
  • Mirrored volume can be accessed down any path
  • Writes, SCSI reservations etc are mirrored across to other mirror plex(es)
  • If a mirror plex fails VMware simply uses another path
    • Which is actually a path to a SVA which is still functioning
  • Can withstand VMware server failure or DAS storage failure
  • Mirrored volume appears to VMware as a regular iSCSI volume with multiple paths
  • Active-active mirror
  • I/O can go down any path to any plex
  • If a physical server fails, VMware simply sees a path failure and fails over to an alternate path, which is actually a path to the other side of the mirror, running on a difference server