The Cloud Makes It Rain Money: The Business in Cloud Computing - Deploying and Managing a Cloud Infrastructure - page 366

However, the challenging aspect of using cloud-based backups is the recovery proce-
dure. With bandwidth limitations and a lot of data to recover, getting data restored on site
within defined RTOs can be tough. To minimize the recovery time, an option is to use local
backups that are copied to the cloud as well. In this way, the backup in the cloud becomes a
secondary offsite backup.
Service features such as compression and data deduplication make data restores from the
cloud to the on-premises infrastructure a viable option. Another practical option is to back
up and restore applications and data using VMs or virtual disk images (VDIs). Prestaging
disaster recovery VMs and keeping them relatively up-to-date through scheduled restores
can help in cases where a string of failures strike. This can help in a situation where aggres-
sive RTOs need to be met.
Replication is the key for applications that require quick recovery time and recovery point
objectives (RPOs) while supporting application awareness. An RPO ( http://en.wikipedia
.org/wiki/Recovery_point_objective ) can be defined as maximum tolerable time period
for which data might not be available from an IT service because of a major disruptive event.
This means that the system design should take into account a limit to work toward. The
system design should also include off-site mirrored backups, which must be continuously
maintained.
There are two types of replication procedures ( http://en.wikipedia.org/wiki/
Replication_(computing) ):
Synchronous (or Eager) Replication Synchronous or eager replication focuses on prevent-
ing conflicts. For example, if a record is simultaneously changed on two or more nodes, an
eager replication system would be able to detect the conflict. The system will then abort
one of the transactions before committing the other.
Asynchronous (or Lazy) Replication Asynchronous or lazy solutions perform conflict
resolution. For example, if a record is simultaneously changed on two or more nodes, a lazy
replication system would first allow both transactions to perform commit. The system will
then execute a conflict resolution resynchronization phase from periodically. The conflict
resolution resynchronization phase could be based on a number of factors, such as transac-
tion timestamp or node hierarchy.
Backup Sites and Geographical Diversity
Recovery can take place only if there is a location from which it can take place. This loca-
tion is known as a backup site. In the event of a disaster, a backup site is both where an
organization's data center will be re-created and the location from which that organization
will resume to operate for the entire length of the disaster.
There are three different types of backup sites, cold, warm, and hot, referring to the
effort required to begin operations at the backup site:
Cold Backup A cold backup site is something akin to a properly configured space in an
office building. Everything required to restore service to users must be procured and delivered
to the site before the recovery process can begin. The delay to reach full operations can be
substantial. However, this is the least-expensive backup site solution.