It is no secret that IT infrastructure of US government agencies, sprawling and disorderly as it reportedly is, can be more efficient if agencies integrate an appropriate form of cloud computing into their IT strategy.

Some agencies are exploring the model and some have already implemented some type of cloud computing. Of them, National Aeronautics and Space Administration is at the forefront, developing one of the most unique and innovative systems in the world: the Nebula Cloud Computing Platform.

Several projects making up the program are currently in various states of completion. Applications already in production include USAspending.gov, a US federal government Web portal that publishes information on government spending, and Microsoft’s WorldWide Telescope, an application that provides public access to images of space from some of the world’s most powerful telescopes.

The program was originally called NASA.net and was primarily meant to support Web-application development framework on behalf of science missions across the agency, said Joshua McKenty, who contracts with NASA as Nebula’s chief architect. “In order to deliver that, we realized that we actually needed some elastic infrastructure of our own.” No commercial vendor at the time could provide elastic infrastructure as a service for that purpose and Nebula became a side project to NASA.net.

Economics was a key impetus for the program. “Data center space has been a major driver at Ames. We’ve been running out of it.” Increasing capacity, while centralizing and increasing efficiency of the infrastructure were the main drivers.

NASA currently has 4,000-6,000 Web sites, spread out between data centers and individual servers. Some of these sporadically experience large traffic spikes when certain projects the agency is involved in attract a lot of public attention and the cloud model provides an efficient way to handle these spikes.

Once fully functional, Nebula will provide Infrastructure-as-a-Service, Platform-as-a- Service, Software-as-a-Service and Databaseas- a-Service, according to the program’s Web site

WHO WILL USE NEBULA?

The platform will serve everyone who works at NASA. Its services will be part of basic entitlements the agency’s employees enjoy. “If they are entitled to have a desk and a computer and a phone, they’ll be entitled to have Nebula services,” McKenty said. There will, however, be charge-back mechanisms for those who use its services more extensively than others, he added.

Discussions are also underway with the office of US Federal CIO Vivek Kundra – a vocal supporter of government clouds – about the possibility of Nebula serving agencies outside of NASA. While members of the Nebula team are wary of the prospect, likely to turn into a governance nightmare, they will not be in the position to decline a request from the White House to share the resource if such a request does finally come.

“Providing infrastructure to a broader number of government agencies gets very complicated,” McKenty said. “We have a small team and a small budget and we have more than enough customers waiting inside NASA.”

Within the agency, there are three types of future Nebula users. The first group consists of employees that do not have budgets for any substantial IT resources and currently have to come up with creative solutions, like storing data on disks. “They are very excited about Nebula because it gives them the capability to do more.”

The second category includes users that have budget for IT hardware but only need it for finite periods at a time for certain projects, after which the gear rarely gets reused. A cloudbased resource is a perfect solution for them.

Employees already using the agency’s data center resources comprise the third group. Leadership at the existing NASA data centers is already making consolidation and virtualization efforts and users from the third group are looking at cloud computing “not so much for efficiency but for some of the security and management advantages.”

For the Nebula team, serving the first two groups is a priority.

A NEBULOUS ARCHITECTURE

The container housing most Nebula infrastructure was made by Verari, a San Diego, California-based IT vendor. The box’s total power draw is 350 kW. It has about 15 equipment racks inside, with an inrack cooling set-up. The unit is connected to external chillers. There are several compute clusters inside, made by a variety of vendors. Some of them NASA has purchased and some are in trial mode, since engineers are continuously evaluating various configurations, looking for the optimal solution.

Gear already purchased includes one Cisco Unified Computing System cluster, two Silicon Mechanics clusters and one cluster of Verari blade servers. HP and NetApp gear is currently deployed in evaluation mode and the team is looking at testing Dell and Fujitsu equipment in the near future.

Excluding evaluation gear, Nebula has about 1,000 processor cores in production today. There is about ¾ of one petabyte of storage currently deployed.

The reason for having a wide variety of equipment is that the team is still in the process of pin-pointing the proper combination of CPU, RAM and disk resources in a single server. “You can oversubscribe storage and CPU, but it’s very hard to oversubscribe RAM,” McKenty said. “I have yet to find the perfect vendor that provides (exactly) what we need.

“We’re trying to build a platform that is opensource and, for the most part … it’s vendoragnostic. I don’t care who the vendor is, as long as it works.” Once completed, the Nebula architecture will be shared with others who wish to build a similar platform of their own.

All networking in Nebula is 10G Ethernet, which is not considered “bleeding-edge” but, from McKenty’s observations, remains short of being well-understood. In his team’s experience, there was not a single vendor whose 10G Ethernet device worked perfectly out-of-the-box. The architecture uses core network switches supplied by Cisco and network cards by Intel and QLogic.

The platform is using KVM virtualization technology. The team originally started with Citrix’s Xen hypervisor, but it had performance issues on the network layer the engineers could not work around. These problems were more easily solved with KVM.

Building Nebula has been a learning process for both NASA engineers and the vendors. Problems the team encountered in building the system were nothing like the problems they or the vendors had expected. Once the issues were identified, however, they were not too hard for the vendors to fix.

One of the biggest challenges was finding a way to shut hardware down when it was not needed and to bring it back up quickly enough to provide resource elasticity and realize efficiencies that are at the heart of the IaaS delivery model. Factors that prevented seemingly similar server designs from booting up quickly enough were in many cases unpredictable, such as BIOS configuration or the type of hard-drive installed.

Nebula’s approach to redundancy is different from the traditional data-center uptime philosophy. It does not strive for maximum redundancy possible at the individualcomponent level, aiming instead do distribute any particular application across enough locations so that the application as a whole stays up even when an infrastructure component fails. They are not there yet.

A second Nebula set-up is being developed on the East Coast of the US, which will give the platform the kind of redundant infrastructure it is aiming to have. More containers are sure to be built and brought online within the next several years. No exact strategy for future capacity increases has been worked out yet.

Besides the primary container, parts of the Nebula infrastructure are deployed at the Ames data center it is next to and another part is hosted at a facility in southern California. The platform will use less than its total capacity once fully launched, but the extra equipment will be switched off and then brought online gradually, as needed.

FOCUS ON MOBILITY

Besides reducing cost and increasing efficiency, mobility played a key role in choosing the containerized approach. At first, the idea was to be able to put the container on one flat-bed truck and the chillers on another flat-bed truck to relocate the system at any time necessary. When the idea was tested, the team was able to set up the entire system in a single day. Still, McKenty is not sure whether that approach will be the permanent solution. Having a highly mobile cloud will be beneficial to NASA because it often partners with universities around the country on research projects and space missions. Usually, there will be several years of development and the mission itself will last six-12 months, during which enormous amounts of data will be collected. “At the end, that data needs to become part of the long-term storage and sharing apparatus of NASA,” McKenty said. Instead of transferring data over the Internet, which has in the past taken about one month for a single mission, driving the Nebula container to the institution where the data is stored and transferring it into the cloud on-site would be faster and easier.

SO, WHAT KIND OF A CLOUD WILL NEBULA BE?

While not fitting perfectly any of the generally accepted descriptions of different types of clouds, Nebula will most closely match that of a “hybrid cloud.” In addition to providing services to NASA internally, the team is working on enabling its users to utilize resources deployed in a commercial vendor’s public cloud when appropriate, such as the times when data requested is public and does not have to be deployed on the private infrastructure.

As previously mentioned, the platform’s architecture will be made public once completed so anyone can build a similar system for themselves. The program’s trialand- error process has been a steep learning curve for NASA engineers and all the major IT vendors, pushing cloud computing ahead technologically in a big way, from which the entire IT industry stands to benefit.