Mix and Match
BuildFax's Emison still runs most of his production systems on Amazon's cloud infrastructure. But he has shifted some to Google Compute Engine and hopes to move more, while keeping enough on Amazon to provide failover for business continuity.
He's using cloud management software from RightScale to, among other things, create servers from standard templates, and to deploy and monitor them on multiple clouds without using a separate API for each. He estimates that BuildFax saves at least $150,000 per year by not having to hire server and network administrators.
But he has also realized that "with a public cloud, you're paying for good security, redundant network connections, redundant power" that BuildFax doesn't need for internal tasks such as data cleansing and indexing. It can reduce its general server budget by 60% to 65% by doing that work in-house, using RightScale to quickly move those workloads to public clouds if needed.
Sonian uses IaaS to run the all-important search capability it provides to users, but it uses a SaaS offering for its less critical relational database. While Amazon has a PaaS search option, Arnette says he chose IaaS "because we would have more control over the design of that very critical feature."
Robert Jenkins, CTO and co-founder of cloud vendor CloudSigma, says his company has an e-commerce customer that "keeps the core database in their own infrastructure" and spins up public cloud servers when needed. He notes that "we could probably [also] move their data into our cloud effectively," but the customer is too protective of its company data to take that step.
Moving to a multicloud world requires not only thinking through risk-benefit trade-offs, but also changing application architectures and IT processes.
Best Practices
In-house data centers include servers, storage and networks that provide certain levels of performance and reliability. One reason IaaS is so inexpensive and flexible is that many customers share a vendor's resources, so performance isn't guaranteed -- it can rise and fall based on the demand from other customers. That could happen if, for example, "the guy sharing my server [is] streaming music and hogging the server," says Momchil Michailov, CEO and co-founder of data management vendor Sanbolic.
If latency causes an IaaS application to fail, the system should alert an administrator and resume work when another server comes online without manual intervention or lost transactions. Emison says you should "assume your public cloud servers are going to fail" and architect systems so that it's easy to provision another from the "infinite pool" available in public or private clouds. Using RightScale, Emison says BuildFax can automatically create a cloud-based server that can, for example, access a file of building permit data, "extract all the data into a text file, load it back up into our central data store and self-terminate. If it dies in the middle [of the process], it spins up again."
Using a service-oriented architecture abstracts services from the hardware, creating an asynchronous environment that's less dependent on specific performance levels, says Pepple.
Many observers stress the need to avoid a single point of failure -- a need highlighted by well-publicized outages at Amazon Web Services. In a recent blog post, Todd McKinnon, CEO of Okta, said he chose IaaS for his company's cloud-based identity management service because it forced him to avoid "supposedly 'highly available' components like expensive load balancers and shared-state, clustered databases." However, he says, none of the IaaS providers he has worked with can tell him whether, for example, two virtual machines share the same physical server. That's why McKinnon says IaaS customers still "need someone in-house" to ensure that their systems are architected correctly.