One of the primary benefits from cloud data operations is the capacity to perform complete backups and to support Continuity of Operations Plans (COOP). COOP recovers operations whenever there is a failure.
In the past data centers were small. Files that needed backup were relatively small, hardly ever exceeding a terabyte. Real-time transactions were rare. High priority processing could be handled by acquiring redundant computing assets that matched the unique configurations of each data center. COOP was managed as a bilateral arrangement. Since hardware and software interoperability across data centers was rare, to restart processing at another side was time consuming. Luckily, data centers were operating at a low level of capacity utilization, which made the insertion of unexpected workloads manageable.
The current DoD environment for COOP cannot use the plans that have been set in place during an era when batch processing was the dominant data center workload. Files are now getting consolidated into cross-functional repositories, often approaching petabytes of data. The urgency of restoring operations is much greater as the processing of information supports a workflow that combines diverse activities.
Desktops and laptops, that used to be self-sustaining, are now completely dependent on the restoration of their screens with sub-second response time. There has been a rapid growth in the number of real-time applications that cannot tolerate delays. What used to be bilateral COOP arrangement is not acceptable any more as DoD is pursuing data server consolidations. The merger of data is based on virtualization of all assets that eliminates much of the spare computer processing capacity.
The current conditions dictate that for rapid fail-over backup data centers must be interoperable. Hardware and software configurations must be able to handle interchangeable files and programs. A failure at any one location must allow for the processing the workloads without interruption at another site. To achieve an assured failover the original and the backup sites must be geographically separate to minimize the effects of natural disasters or of man-caused disruptions. What used to be the favorite COOP plan of loading a station wagon and driving relatively short distances with removable disk packs is not feasible any more. Petabyte data files cannot be moved because they do not exist in isolation. Data center operations are tightly wrapped into a complex of hypervisors, security appliances, emulators, translators and communications devices.
Under fail-over conditions the transfers of data between data centers must be manageable. The affected data centers must be able to exchange large amounts of data over high capacity circuits. Is it possible to start thinking about a COOP arrangement that will operate with fail-overs that are executed instantly over high capacity circuits? How much circuit capacity is required for claiming that two (or more) data centers could be interchangeable?
The following table shows the capacities of different circuits as well as the size of the files that would be have to be transferred for a workable COOP plan:
Transfers of files between small data centers (each with 10 Terabytes of files) would take up to 28 hours using an extremely high capacity circuit (100 MB/Sec). That is not viable.
DoD transaction processing, including sensor data, involves processing of at least one petabyte per date at present, and probably much more in the future. It would take 17 weeks to back up a single petabyte of data from one data center to another even if the highest available circuit capacity of 100 MB/sec is used. That is clearly not viable.
We must therefore conclude that the idea of achieving fail-over backup capabilities by electronic means cannot be included in any COOP plans.
COOP for the cloud environment must be based on multiple data centers that operate in synchronization with identical software, with matching technologies of computing assets and with comparable data center management practices. This does not call for a strict comparability of every application. What matters will be an identity of the DoD private cloud to act as a Platform-as-a-Service utility, with standard Application Processing Interfaces (APIs).
OMB has mandated that for FY12 all new DoD applications will have to cloud implementation options. Rethinking how to organize the DoD cloud environment for COOP will dictate how that can be accomplished.