Many IaaS providers make it difficult to avoid single points of failure because their network architectures don't support clustering or allow more than one IP address to be assigned to components such as virtual servers to allow easy failover, says Michailov.
McKinnon says the geographically diverse data centers typical of IaaS setups make disaster recovery far easier than it is with in-house systems. However, customers should take into account the costs of moving large amounts of data between their own infrastructures and those of their providers. They should also verify that the cloud provider's performance monitoring and reporting systems are consistent with their expectations.
New Tools
Vendors offer various approaches to make it easier to adopt a multicloud strategy and move higher-end applications to the cloud. Most include a software layer to abstract, or hide, the application programming interfaces used by various cloud providers, and to mask the cloud's inherent latency and unpredictability.
Google is among the vendors seeking to provide SAN-like reliability in the cloud. Its recently announced Compute Engine IaaS platform offers network-attached block-level storage that, according to product manager Craig McLuckie, provides the consistency, if not the speed, of more expensive SANs. It also provides what Google claims is 50% more processor power for the money than Amazon and a high-speed network that encrypts customer data in motion and at rest.
Cloud storage provider Zadara Storage offers storage software that's designed to turn standard disk and solid-state drives into "virtual private storage arrays" with the same performance and reliability as SANs, including support for application clustering for high availability, says Noam Shendar, the Irvine, Calif.-based company's vice president of business development.
Sanbolic is offering its Melio data management software and Zadara's Virtual Private Storage Arrays to "deliver enterprise-class storage with high-availability application clustering" on Amazon Web Services and Rackspace. Sanbolic also recently extended support for public clouds to its Sanbolic AppCluster, which provides high availability and scalability for Microsoft SQL Server workloads.
IaaS provider Joyent sells the software it uses in its own data centers to replace virtualization at the level of the physical server, as practiced by the likes of VMware, with a hypervisor at the operating system layer. Steve Tuck, the San Francisco-based company's cloud senior vice president and general manager, says that allows more efficient use of servers, and instant and automatic scaling, or shrinking, of memory and CPU allocation across processor cores as workloads change. It also performs provisioning, migration and management of workloads, he says.
Embotics, a provider of private cloud management systems, targets midsize companies with what it calls "a comprehensive set of virtualization and private cloud management capabilities in a single, integrated software package that can be implemented within one hour."
Another IaaS vendor, TransLattice, uses a protocol that splits the database and associated components such as queuing systems across a network of commodity servers, says CTO Mike Lyle. That protocol, he says, ensures that all the tasks are completed in the proper order, and even handles the locking of transactions of distributed databases.
The Red Hat Hybrid IaaS Solution blurs the boundaries between IaaS and PaaS offerings with cloud-orchestration tools, a self-service portal for users, a virtualization manager, a hypervisor and a guest operating system, as well as the option to add prepaid compute hours from major public cloud providers. Its recently announced Red Hat Storage Server 2.0 will use Red Hat's Gluster file system to replicate and distribute data among multiple public and private data centers, says Bryan Che, senior director for product management and marketing.