Brown CS News

It could be said that Malte Schwarzkopf, who joins Brown CS as Assistant Professor this fall, prefers the long view, and perhaps the wide-angle lens: “Efficiency of datacenter systems is one of the major tracks of my research because the data center stack is important to all of us. Awake or asleep, hundreds of computers are doing things on our behalf, and if they do them inefficiently, that can be expensive and wasteful. How can humanity improve on that? With today’s systems, it’s easy to do a lot of computation at the flick of a switch, but many of these systems are not very discerning in their resource use, and we should aim to do better.”

One way to begin that process, Malte finds, is by excavating the past, and his favorite possessions include a layperson’s guide to technology from 1828. “It illustrates how progress has happened, and creates a connection with people who lived in a very different world,” he says, “and that’s fascinating to me.” On one page, electricity is listed as a curiosity for which no practical use has been found, but one of the book’s owners worked to keep it current by adding annotations and newspaper clippings, including one of the Wright brothers taking to the air for the first time.  

“History helps explain why things are the way they are and why we think the way we do. One great example from computer systems research is that only two decades ago,” Malte explains, “we didn’t have journaling file systems, and I have a vivid memory of Windows 95 crashing during an OS upgrade and running ScanDisk to figure out what had happened and repair the file system. It’s totally different now: all major operating systems use journaling file systems to prevent these problems from arising in the first place. I find it very satisfying to learn from the past when I try to invent new things, and to see what impact computer systems research can have even in, by the dimensions of history, a very short term.”

Most recently, Malte’s research produced Noria, a new system that makes websites more efficient by reducing the redundant computation done at their backend. Noria uses a technique called “dataflow computing” to efficiently precompute the data that the website’s frontend might request. This ensures that reads – which are much more common than writes in these applications – are maximally efficient, improving the load served with the same hardware by 5-74x over common industry practice. In his prior work on datacenter systems, Malte developed cluster schedulers, systems for data science on big data, and a new datacenter network design. His work has received best paper awards at EuroSys 2013 (for the Omega cluster scheduler) and NSDI 2015 (for the QJump datacenter network design).

Schwarzkopf’s introduction to computers came at age seven or eight, in the form of an IBM PC with a 386 processor running at 25 MHz. It was a pretty typical first experience, he says: “I was growing up in Hamburg, and it had been surplused at the insurance company where my father worked, so he brought it home for us. We kept it in the basement and it was a lot of fun.”

Talk of the era before flat screens and smartphones brings us to the second track of Malte’s research. “Unlike in those days,” he says, “our data lives remotely now, and our computers and devices are really just portals. This creates trustworthiness problems about our data being lost, destroyed, or sold that are very much unresolved. Systems play an unappreciated role in this. If we create them in a free-for-all environment, we often end up with bad outcomes, but when we define some parameters, we may get a degree of assurance, a reason to trust.”

By the end of primary school, Malte had started programming in the form of writing macros for Microsoft Office. Self-taught, he eventually became the systems administrator for his high school’s computer system. It was a grassroots effort, he says, with students running the show and building computers themselves, and it was likely the origin for his applied computer science perspective. In the years that followed at the University of Cambridge, Schwarzkopf credits Steven Hand, who later became his PhD advisor, with inspiring his love of systems.

“Broadly speaking,” he says, “I’m interested in all flavors of computer systems, but my recent work has focused on data centers. It’s an interesting area and things are still in flux, with hardware and software always presenting new demands.” Malte’s doctoral research focused on operating systems for large data centers, including work on cluster schedulers, which provide a kind of brain for the data center by deciding which servers should process which work in order to be most efficient.  “Currently, I’m very interested in dataflow systems, as I believe they’re a great abstraction for building scalable infrastructure that is easy to use, but also to reason about.” 

As his research continues, Malte sees Brown as a perfect home for someone who wants to take risks and look ten years down the road instead of being bound to a six-month industry cycle. “And I love teaching,” he says. “I really enjoy seeing students I mentor grow into researchers, and I’m passionate about diversity because we need to move beyond the stereotypes. Even though I work in systems, you don’t need to have been hacking on a kernel in your basement since you were nine. You need to be curious about how things work, keen to build new things, and enjoy tackling intellectual challenges. My wife is a political philosopher, and she provides a lot of input for my research. The best questions come from people who see links and bring a different perspective and are curious – that’s more important than the number of lines of code you have written. We need that broad spectrum.”

The most enjoyable moments of teaching, Malte says, are when neither nor he nor his students know a particular answer, but they approach the problem together. He’s hoping for many such moments in his new course (CSCI 2390 Privacy-Conscious Computer Systems), which focuses on systems that protect user data by design through enforcing predefined rules. “We need that kind of two-way learning because so many of these problems are unsolved, even for simple web apps. We need to find ways for companies to run back-end systems that won’t harm users, and a lot of theoretical work has been done in this area, but many of the concepts are impractical and will never be deployed. Legislators are waking up, and they’re asking for system designs that better protect sensitive information at a cost where people can use them.”

Ultimately, Schwarzkopf hopes that he and his students will respond to that need with feasible, affordable, and efficient solutions. The way to build them, he says, is by challenging each other to be better engineers, and more responsible ones: “Building practical systems is in my DNA, and I want to keep doing that with my students, making things that work and are worthy of trust and have some certainty of becoming widely used.” 

Perhaps someday they’ll be able to write their own names among the marginalia of that book of his. “As academic researchers,” Malte says, “we’re part of history. We’re telling our future selves why they do what they do as scientists, as creative humans.”