29.12.2014 Views

Magellan Final Report - Office of Science - U.S. Department of Energy

Magellan Final Report - Office of Science - U.S. Department of Energy

Magellan Final Report - Office of Science - U.S. Department of Energy

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

<strong>Magellan</strong> <strong>Final</strong> <strong>Report</strong><br />

cloud environments. Similarly, I/O intensive applications take a substantial hit when run inside virtual<br />

environments.<br />

• Cloud programming models such as MapReduce and the resulting ecosystem show promise for addressing<br />

the needs <strong>of</strong> many data-intensive and high-throughput scientific applications. However, current<br />

tools have gaps for scientific applications. The MapReduce model emphasizes the data locality<br />

and fault tolerance that are important in large systems. Thus there is a need for tools that provide<br />

MapReduce implementations which are tuned for scientific applications.<br />

• Current cloud tools do not provide an out-<strong>of</strong>-box solution to address application needs. There is<br />

significant design and programming required to manage the data and workflows in these environments.<br />

Virtual machine environments require users to configure and create their s<strong>of</strong>tware images with all<br />

necessary packages. Scientific groups will also need to maintain these images with security patches and<br />

application updates. There exist a number <strong>of</strong> performance, reliability, and portability challenges with<br />

cloud images that users must consider carefully. There are limited user-side tools available today to<br />

manage cloud environments.<br />

• One way clouds can achieve cost efficiency is though consolidation <strong>of</strong> resources and higher average<br />

utilization. DOE Centers already consolidate workloads from different scientific domains and have<br />

high average utilization, typically greater than 90%. Even with conservative cost analysis, we show<br />

how two DOE Centers are more cost-efficient than private clouds.<br />

As noted in the final point, a key benefit <strong>of</strong> clouds is the consolidation <strong>of</strong> resources. This typically leads to<br />

higher utilization, improved operational efficiency, and lower acquisition cost through increased purchasing<br />

power. If one looks across the scientific computing landscape within DOE there are variety <strong>of</strong> models for how<br />

scientists access computing resources. These cover the full range <strong>of</strong> consolidation and utilization scales. At<br />

one end <strong>of</strong> the spectrum is the small group or departmental cluster. These systems are <strong>of</strong>ten under-utilized<br />

and represent the best opportunity to achieve better efficiency. Many <strong>of</strong> the DOE National Laboratories<br />

have already taken efforts to consolidate these resources into institutional clusters operating under a variety<br />

<strong>of</strong> business models (institutionally funded, buy-in/condo, etc.). In many ways, these systems act as private<br />

clouds tuned for scientific applications and effectively achieve many <strong>of</strong> the benefits <strong>of</strong> cloud computing. DOE<br />

HPC centers provide the next level consolidation, since these facilities serve users from many institutions<br />

and scientific domains. This level <strong>of</strong> consolidation is one reason why many <strong>of</strong> the DOE HPC centers operate<br />

at high levels <strong>of</strong> utilization.<br />

Clouds have certain features that are attractive for scientific groups needing support for on-demand access<br />

to resources, sudden surges in resource needs, customized environments, periodic predictable resource needs<br />

(e.g., monthly processing <strong>of</strong> genome data, nightly processing <strong>of</strong> telescope data), or unpredictable events such<br />

as computing for disaster recovery. Cloud services essentially provide a differentiated service model that<br />

can cater to these diverse needs, allowing users to get a virtual private cluster with a certain guaranteed<br />

level <strong>of</strong> service. Clouds are also attractive to high-throughput and data-intensive workloads that do not fit<br />

in current-day scheduling and allocation policies at supercomputing centers. DOE labs and centers should<br />

consider adopting and integrating features <strong>of</strong> cloud computing into their operations in order to support<br />

more diverse workloads and further enable scientific discovery. This includes mechanisms to support more<br />

customized environments, but also methods <strong>of</strong> providing more on-demand access to cycles. This could be<br />

achieved by: a) maintaining idle hardware at additional costs to satisfy potential future requests, b) sharing<br />

cores/nodes typically at a performance cost to the user, and c) utilizing different scheduling policies such<br />

as preemption. Providing these capabilities would address many <strong>of</strong> the motivations that lead scientists<br />

to consider cloud computing while still preserving the benefits <strong>of</strong> typical HPC systems which are already<br />

optimized for scientific applications.<br />

Cloud computing is essentially a business model that emphasizes on-demand access to resources and<br />

cost-savings through consolidation <strong>of</strong> resources. Overall, whether cloud computing is suitable for a certain<br />

application, science group, or community depends on a number <strong>of</strong> factors. This was noted in NIST’s draft<br />

document on Cloud Computing, “Cloud Computing Synopsis and Recommendations”, which stated,<br />

130

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!