Analytics Environments on Demand, or AEoD, is an emerging service that provides remotely-accessed, domain specific, scalable computational environments to UC Berkeley researchers. For researchers looking to work with bigger data sets or accomplish larger tasks than their laptops or desktops can reasonably accommodate, including those who are considering future applications of even more robust computing infrastructure (e.g., High Performance Computing), AEoD’s familiar interface and ease of use offer an effective first step towards more computationally intensive inquiry.
Currently in a second round of service experiments with test users at the Goldman School of Public Policy (GSPP), the Haas School of Business, D-Lab, and the Digital Humanities Berkeley Summer Institute, AEoD will give researchers compute power and memory well beyond the capabilities of their personal computers, yet will be accessible from the web browser of a personal device, like a laptop or tablet. The virtual tool promises to have a wide impact across the campus, meeting specific desktop and application needs of many diverse scholars. The service is spearheaded by IT Architect Jason Christopher, a member of the Berkeley Research Computing (BRC) team, with whom we spoke regarding AEoD.
The Service: What is AEoD?
Suppose you use a desktop computer to access, record, transform, or analyze data for your research. This is an Analytics Environment! The analytics are the set of software used to work with your research data, and the environment is the operating system (such as Windows, MacOS, or Linux) in which you run analytic software. This compute platform is domain specific: it includes the tools you selected to fit your research. The analytics software might include statistics packages, such as STATA or SPSS; software that manipulates and presents geospatial data, such as ArcGIS; or software that performs Natural Language Processing (NLP), such as the packages written and maintained by the Stanford NLP Group. There are, however, limitations to the work you can do on this desktop platform. For example, your computer may not have adequate CPU power or RAM (memory) to complete a large, complex task within a reasonable number of hours, and running such a task may slow down or preempt other applications (email, messaging, word-processing, etc.). So what do you do?
One possibility is to switch from your desktop to the AE or Analytics Environments, in AEoD. Leveraging existing support for Citrix servers administered by Information Services and Technology’s Platform Infrastructure group, BRC will provide analytics environments that match researchers’ desktops, except that they are spun up as virtual machines (VMs) with as much as 192 GBs of RAM and up to 16 CPU cores, depending on need. Analytic tasks can run faster with these more substantial resources; and by ‘outsourcing’ tasks from a researcher’s personal computer, that desktop (or laptop) is freed up to perform other functions, from checking e-mail to polishing a grant proposal -- or to be shut down or go offline.
Christopher emphasizes that “ease of use is a guiding design principle” for the AEoD service. Hence oD, or on Demand, the second component of the service. A virtual desktop instantiated on remote server hardware, an AEoD VM can be accessed like any web application, from a web browser. Moreover, the virtualized environment is designed to be familiar to the user. As Christopher notes, “Why teach researchers how to use a whole new model of computing, or a new operating system? We want people to be able to compute in the same way, using the model they’re already familiar with, but they’ll be able to compute on much larger resources.” The AEoD team is now exploring an even more streamlined option: providing single applications to researchers via a VM platform, removing unnecessary OS overhead, and “giving them the minimum that they need, but meeting all of their needs.” For researchers with an “occasional need to run STATA, or SPSS, but for jobs that would require 50, 60, 80 GB of RAM,” Christopher continues, “this single application delivery may be a great solution.”
Implementation and Testing
In a series of second-round experiments, the AEoD team at BRC has partnered with several campus departments, including the Haas School of Business, the Goldman School of Public Policy, and the D-Lab. The goal, Christopher explains, is “to test out this infrastructure we’ve put together, and see how well it will work under usage of 10, 20, 30 or more concurrent users.” According to Christopher, the Haas School of Business is interested in the service because many researchers at the school are using STATA on Windows OS research desktops maintained by the school. The data they work with is typically very large, so if the promise of 80 plus GB of RAM in ÆoD VMs proves viable, Haas may be able to consider scaling back or eliminating their existing service.
AEoD is expected to provide a solution to issues of resource allocation and management that Haas researchers experience with their existing service. The issue surfaces when a researcher whose compute job requires a lot of memory begins to shut out other users already running tasks in the virtual workstation. As she soaks up all of the available RAM, other users’ jobs slow down, and in the worst cases are stopped altogether. Solving this problem requires “resource isolation” between virtual environments running on the same servers. The policy issue that must be worked out, Christopher explains, is, “How will resources be allocated to groups with multiple people, when one person’s job requires more memory than another?”
Christopher and the AEoD team are working to implement resource isolation through the current, second round of experiments. They are exploring several solutions, but any adopted solution will be augmented by “support from BRC consultants and local IT people [in the test user’s department] who can raise awareness among researchers about resource isolation and train users to better manage their jobs,” Christopher says.
Looking Forward
As ÆoD experimentation continues, BRC is preparing to extend a limited service to early adopters. Christopher points out, “You have to get people using a service to see how it will work and fix things that come up, but there is a lot of promising interest in the abilities of AEoD.” Keep an eye on Research IT News for future articles highlighting campus researchers who are actively benefiting from this emerging service!
For more information, contact research-it@berkeley.edu.