Academic Collaboration: Accelerating the Practical Application of Research

By Jeff Parkhurst, Ph.D., Intel Labs

The Intel Science and Technology Center Program is a series of research collaborations that pairs Intel with U.S. universities to identify and prototype revolutionary technology opportunities. The centers are designed to encourage closer collaboration among academic thought leaders in essential technology areas. Jeff Parkhurst is the Intel Program Director for the Big Data and Cloud Computing centers. They are working on new technology infrastructure essential for Big Data: a new ecosystem for Big Data.

In 2011, Intel took a new approach to funding university research. Instead of funding research and working at arm’s length with university professors, we decided to try a new paradigm. Enter the concept of the Intel Science and Technology Center, which differs in a number of ways from the traditional way of working with universities.

The key to the success of the program involves working closely with both professors and students through assigning technology stakeholders and embedding full-time researchers on campus who co-author papers as well as co-advise students. Additionally, we create a community under the center umbrella consisting of professors from a variety of universities providing diverse expertise. We even encourage each center to collaborate informally and practically with the other centers with the end goal of addressing large problems that could not be solved by any one university or any one center. We are always looking for opportunities to provide not only hardware platforms but also real-world, practical datasets so that professors and students don’t have to work with manufactured data, which has limited value. Finally, all of this research is open intellectual property (IP).

To our delight, Intel and the centers are collaborating more widely, more deeply and more creatively than we ever expected. Collectively, we are achieving new levels of industry-university collaboration, allowing people from multiple institutions and multiple disciplines to help each other succeed. And the best part is that the collaboration keeps getting better, as people see new paths to research exploration.

From this foundation, we are seeing a variety of Intel technologists formally collaborate with the professors and students who are part of the center, and I continue to get additional inquiries as to the research and where new engagements might be possible. Occasionally, we get multiple individuals from different departments within Intel seeking to engage on similar topics of research. Both for messaging and organizational purposes, we have started the concept of what I call research working groups. These groups consist of multiple technology stakeholders across Intel and multiple professors sometimes spanning universities.

Collectively, we are achieving new levels of industry-university collaboration, allowing people from multiple institutions and multiple disciplines to help each other succeed. And the best part is that the collaboration keeps getting better, as people see new paths to research exploration.

I’d like to mention just two examples.

Expanding Visualization Research with a “Live” Retail Database

We have a research working group focused on Big Data visualization. This group consists of people from our Data Center Group and our Internet of Things Group on the Intel side working with Professors Bill Howe and Jeffrey Heer of the University of Washington.

Big Data visualization is of high interest vs. more of the standard query-in/table-out methods. In the latter case, the answers you get depend on the questions asked. Additionally, with large amounts of data, it can become quite cumbersome to wade in and find information you are seeking.

Compared to tabularized data, visualization enables you to process large amounts of data more naturally and extract additional information beyond the initial query.

To create more value in this particular research, Intel provides our university collaborators with a real-world retail dataset from which to work.

Visualization is especially helpful when one is trying to understand correlation across items with varied interdependencies. Consider the situation where you have four or five variables and you want to understand their correlation. For instance, in retail, buying a particular product might be tied to seasonal buying, item placement in the store, sales or promotional events, product dependencies, day of the week, etc. Unique types of visualizations are required to properly display all this information in consumable form.

At some point in time, this process of selecting the right visualization needs to be automated. This requires intelligent selection leveraging machine learning principles as well as understanding the historical information about dataset queries and usage. This is also being incorporated into the research with the hope that we can develop a better automated system or one that can help the user intelligently choose which visualization path to follow.

This particular working group meets regularly at a six-week cadence.

Non-Volatile Memory Working Group

The other working group is the non-volatile memory research working group. It meets every two months. On the academic side are people from Carnegie Mellon and Brown University. On the Intel side are technology stakeholders from Intel Labs, our Non-Volatile Memory Systems Group, and our Software Services Group.

With memory capacity increasing rapidly, we are seeing traditional applications that used to reside on disk now being moved to main memory. These would include in-memory databases. However, fault tolerance must be considered in the event of an unanticipated shutdown and with traditional main memory DRAM, all your data would be lost. Many packages include periodic writes to disk and logging as a way to recover in these cases.

Emerging Non-Volatile Memory (NVM) technologies provide multiple advantages over existing DRAM-based main memory such as large capacities and load/store accessible persistent storage of data.  However, NVM memory technologies have different latency and bandwidth characteristics than DRAM. While some of this can be addressed through careful hardware system design, it also exposes new opportunities and tradeoffs in application design and implementation to optimize for future NVM-based main memory systems. The research we are funding at the center is looking at these tradeoffs and discovering the best use cases for this technology.

Intel is providing a hardware emulator of two types of systems: one is a pure bank of NVM while the other is a mixture of DRAM with some NVM. We provide guidance on how to use the hardware; we also provide feedback on university research results.

Both of these projects combine an academic expertise with technology stakeholders from industry focusing on real-life problems that people want to solve. Research working groups are taking an out-of-the-box approach to Big Data, a different look at pragmatic problems. This approach, we hope, will lead to additional usage models for these nascent technologies.

Research working groups are taking an out-of-the-box approach to Big Data, a different look at pragmatic problems. This approach, we hope, will lead to additional usage models for these nascent technologies.

About Working Groups in General

As I mentioned, not all collaboration need involve a working group. However, where we have them instituted, they have become quite successful.

On the Intel side, it means multiple groups from Intel coming together in one meeting and providing a consistent, prioritized and holistic message.  On the university side, it helps continue the theme of building communities, allowing them to solve much larger problems than they would have solved individually.

 

This entry was posted in Big Data Applications, Computer Architecture, DBMS, ISTC for Big Data Blog, Visualizing Big Data and tagged , , , , , , . Bookmark the permalink.

Leave A Reply

Your email address will not be published. Required fields are marked *


9 × = sixty three