Widespread availability of high-throughput DNA sequencing instruments, and the development of novel techniques based on sequencing, provides a potentially very valuable resource for researchers in genomics. However, transforming sequence data into biologically meaningful information requires sophisticated computational infrastructure and support. The size of the required computational infrastructure is outpacing what many labs and even universities are able to support. In addition, the setup and maintenance associated with a computational infrastructure presents significant problems for individual investigators and small labs that may not have the necessary informatics support.

Fortunately, a computational model – cloud computing [1] – has recently emerged and is ideally suited for the analysis of large-scale sequence data. In this model, computation and storage exist as virtual resources in remote datacenters, and can be dynamically allocated and released as needed. However, cloud computing resources are not yet suitable for immediate “as is” use by experimental biologists. In the current model, cloud resources are acquired as independent, stripped-down units that must first be customized for the intended use. They then must be configured to work in unison and a mechanism must be provided for the data uploaded and analyzed on those resources to persist beyond the life of a (usually transient) cloud compute resource.

To date, there are several projects and solutions in the context of high throughput sequencing and bioinformatics in general that utilize cloud computing to deliver the computational capacity and on-demand scalability (e.g., Crossbow [2], CloudBurst [3], RSD-cloud [4], Myrna [5]). However, these projects primarily target specific problems and provide custom solutions for a given tool or methodology. Although very valuable, such tools provide minimal help for researchers wanting to compose a variety of tools into an analysis pipeline or for researchers that want to use their own tools when utilizing cloud computing resources. Thus, there is a need for a flexible solution that can be utilized in a variety of scenarios and that provides support for customization.

This paper introduces CloudMan from the Galaxy Project [6], an integrated solution that leverages existing tools and packages by providing a generic method for utilizing those tools on cloud resources and abstracting out low-level informatics details. This solution handles all of the intricacies of cloud computing resource acquisition, configuration, and scaling to deliver a personal compute cluster in a matter of minutes. All interaction with Galaxy CloudMan and the associated cloud cluster management is performed through a web based user interface and requires no computational expertise. The deployed cluster comes preconfigured with all of the bioinformatics packages available in the NERC Bio-Linux workstation (version 6) [7] as well as a range of NGS tools available within the Galaxy framework [8]. Moreover, the process of tool deployment is fully automated and decoupled from the base machine image, making it possible to very simply add additional tools to an individual cloud cluster instantiation.

To keep up with the growth of data being produced in life sciences and the accompanying computational demand, there is a need for increased access to computational resources. Cloud computing offers access to such resources but still makes it difficult to create complex deployments of useful standalone infrastructures. This is especially cumbersome for individuals and small labs that lack informatics support to fully harness this general-purpose infrastructure.

The default version of Galaxy CloudMan can be used “as-is” to support creation and control of fully functional compute clusters on cloud resources; it supports a broad range of bioinformatics tools and it makes it possible to add additional tools with little effort. The process of tool deployment is completely automated and well documented, making it reproducible in other environments. Overall, the system is simple to use and it targets individual researchers so they can gain access to the computational resources they need without requiring support from skilled bioinformatics personnel.

The source code for the entire project is available under the MIT licence and is available from http://bitbucket.org/galaxy/cloudman/. Documentation and detailed instructions on how to use Galaxy CloudMan are also available from http://usegalaxy.org/cloud.

To deliver a ready-to-use compute infrastructure to individual researchers, we decided to build CloudMan as a modular framework that can be used “as is” but also supports easy customization. In keeping with this goal, we have embedded Galaxy CloudMan on top of the Bio-Linux workstation machine image and integrated it with Galaxy. This delivers a broad range of bioinformatics tools directly to CloudMan users.

To support the evolution of the tools and the project as a whole but minimize disruption of end user’s routine, the Galaxy CloudMan has been built with the modularity of the cloud infrastructure in mind. Specifically, we have the domain specific tools separate from the core components that enable CloudMan to operate. In the context of EC2, this is achieved by having a “thin” machine image that contains only the essential bootstrapping scripts and a well-defined instance boot process, while the desired tools reside on external data volumes (i.e., EBS volumes). Such an approach enables the same machine image to be used for a variety of scenarios and to evolve over time because it can be contextualized at runtime (i.e., prepare and deploy an otherwise generic instance) [17]. This model enables us to update the functionality of CloudMan without requiring the users to alter their routine; they continue to use the infrastructure as they are accustomed to and still have the option of updating the functionality offered by CloudMan (Figure 3).

By having domain specific tools on external data volumes, it is easy to modify the content of those volumes, and thus the availability of underlying tools, without the need to modify and rebuild the machine image. This is convenient for end users as they can focus on using the tools while the CloudMan developers can focus on ensuring the infrastructure behaves as desired. Specifically, in order to update a tools data volume, a user instantiates a cloud cluster, modifies the contents of the tools data volume by installing the desired tools (performed just like on any other machine and also showcased by our tool deployment script), creates a snapshot of the modified data volume (done automatically through the web UI), and points their cluster setup to use the newly created data volume snapshot (done manually through a web UI). As a result, the functionality and the architecture of CloudMan can easily be adjusted to the computing needs of individual researchers and different domains. This makes it possible for end users to add their own tools to a CloudMan cluster instance.