All about Clouds: In the Sky and in Computing

Michael Zink, professor of computer and electrical engineering, came to UMass Amherst in 2004 to work for CASA, the Center for Collaborative Adaptive Sensing of the Atmosphere. Established in 2003 by the National Science Foundation (NSF), CASA has been a partnership among academia, industry, and government that engineers revolutionary weather-sensing networks. These networks save lives and property by detecting the region of the lower atmosphere below conventional radar coverage—mapping storms, winds, rain, and other hazards. Building on his work in multimedia streaming, distributed systems, and networking, Zink now leads several major NSF-funded research projects in new cloud computing, and is a principal investigator of the Massachusetts Open Cloud (MOC). In addition, he serves as co-director of the new Paros Center for Atmospheric Research and was named the first Paros Professor of Geophysical Sensing in 2022.

CASA just had its 20th anniversary. What has been its greatest accomplishment?

We proved that short-wavelength technology could work for short-term, low-to-the-ground weather forecasts. A lot of people thought it couldn’t be done.

What’s next for CASA?

As CASA transitions to the Paros Center for Atmospheric Research [see Paros Center pg. 24] we will broaden our mission in terms of research for improved atmospheric modeling, weather forecasts, and weather warning. We’ll work with other institutions on forecasting natural events, including hurricanes, tsunamis, and earthquakes, using sensors as well as radar.

What have you learned from your work with CASA?

The importance of the human component. An epiphany for me as an engineer is that you have to create information that people use in the right way. We could come up with all kinds of fancy technology and algorithms to predict flooding, but if people continue to drive through flooded areas, something is wrong.

Can you draw the connection between your work with CASA and your recent research in cloud computing?

You have clouds in the sky and clouds in computing, but it’s not only the words. In both cases we’re networking data and making sure data is going to the right places and can be used for efficient computation. Weather modeling and weather forecasting have been on the cutting edge of high-performance computing; many of the world’s supercomputers belong to national weather services.

We can use cloud computing to make compute needs and compute demands more efficient. For example, in forecasting the computing load changes with the weather. On a perfect sunny day, there’s not too much to compute. On a day where the weather is stormy, you need a lot more computing power. In the old model you work with a cluster that costs a lot of money; at times it may be idle and at other times it may be overloaded. It’s not elastic and doesn’t adapt to your needs. However, with cloud computing you can allocate compute power in a dynamic fashion.

You’ve been working with test beds since 2014. Why is there a need for experimental platforms for cloud computing?

Testbeds allow diverse communities to exploit new cloud technologies. They democratize cloud- computing research and allowing increased collaboration between the research and open-source communities. This is research that could change the operation of the cloud itself.

What are some of today’s challenges in cloud computing?

Two of the biggest challenges are security and storage. How can you improve security when you are moving data among servers? How can you create more efficient storage?

What is the sustainability component of your research?

There are clouds involved in everything these days, from when we Zoom or watch Netflix to scientific research involving huge data sets. We are looking for ways to execute compute jobs more sustainably. For example, in a solar-powered data center if you have a job that can wait, you can delay it until the sun is out. You can also look for ways to combine compute jobs that go to a data center, or run them in a more dynamic fashion. The resulting delay may only be milliseconds, but it conserves significant energy.