Public cloud services might work for applications that don't require ironclad security -- but not in a major teaching hospital where lives are at stake and the integrity and privacy of personal health records must not be compromised.
That, at least, is the opinion of Dr. John Halamka, the CIO of Harvard Medical School and Beth Israel Deaconess Medical Center in Boston. Given the resources available at such a large institution, Halamka determined he could build a better health care cloud on his own.
"Our [internal] cloud strategy is to create private clouds that are more reliable, more secure, and cheaper than public clouds for those applications which require higher levels of availability and privacy," Halamka recently wrote in a blog entry. To that end, he added, Harvard and Beth Israel have essentially transformed public cloud technology into something with a guaranteed service level and compliance with HIPAA as well as Massachusetts data privacy laws.
Health care cloud infrastructure must focus on virtualization, elasticity
The first step in building any cloud infrastructure is virtualization of both servers and storage. This pays off in creating a flexible, scalable computing and storage resource. In addition, virtualization streamlines business continuity and disaster recovery -- both of which are critical in operating a reliable cloud infrastructure.
Server virtualization relies on a hypervisor, such as VMware Inc. vSphere and ESXi or Microsoft Hyper-V in Windows Server 2008 R2, to create multiple logical machines on each physical server. As for the physical servers themselves, the move to a cloud infrastructure might be a good time to swap out old servers in favor of rack-mounted blade servers, which take up less data center space than individual rack-based and tower servers.
A high degree of storage elasticity is required for an effective health care cloud implementation.
A key characteristic of cloud technology is elasticity, or the ability to scale up rapidly by adding computing or storage capacity as needed. Virtualization management software allows the creation of new virtual machines on an as-needed basis should a new application be desired or should a user population increase suddenly.
A high degree of storage elasticity is also required for an effective health care cloud implementation. Virtualized storage is most often implemented with iSCSI or Fibre Channel storage area network (SAN) technologies, which enable the creation of logical storage volumes that can be moved across physical disk drives. Thin provisioning, a technology often associated with SANs, enables the allocation of storage blocks on an as-needed basis. In this way, storage does not have to be allocated for a given purpose in advance, whether or not it is used.
Locking down the health care cloud: multiple layers, locations
Halamka's private health care cloud runs on a 6,000-core blade-based supercomputer that uses one petabyte of distributed storage.
The server is housed in a Harvard facility, is engineered to be highly available, uses grid computing technologies across multiple nationwide locations and is protected by what Halamka has described as "a multi-layered security strategy." Such an approach means, among other things, that data is protected at rest and in transit, that strong user authentication policies are in place and that user and device activity can be monitored throughout the network.
Having a sound disaster recovery site distant from the primary data facility is important to prevent data loss.
Atrius Health, a Massachusetts health care group, uses a backup facility 60 miles from its primary data center. CIO Daniel Moriarty said this makes disaster recovery within Atrius' internal health care cloud environment easier and more efficient. Since virtual server and storage environments allow virtual machines and storage volumes to be easily moved from one set of physical devices to another, it is no big task to replicate a production environment at a remote facility and move workloads there in case of emergency or routine maintenance, he said.
Halamka, meanwhile, uses a similar strategy, placing clinical systems on geographically separated clusters made up of blade-based Linux machines and thin-provisioned storage. Each cluster uses its own Internet connection and is capable of handling terabytes of throughput per day.
Stan Gibson is a Boston-based contributing writer. Let us know what you think about the story; email firstname.lastname@example.org.