Penn Engineering’s Linh Phan is Protecting Cyber-Physical Systems

While the saying “timing is everything” is appropriate in any number of situations, in the world of cyber-physical systems, it’s more literal than most. Such systems are at the heart of technology where the difference of a second is the difference between life and death.

Basically tiny, dedicated computers that control real-world objects, cyber-physical systems, or CPS, are used to regulate everyday devices such as insulin pumps and automobile airbags, and facilitate large manufacturing systems in factory automation. CPS integrate computing components with physical components, and they are engineered to interact with the physical world through sensors and actuators, ensuring that airbags inflate upon impact or the correct dosages of medication are delivered flawlessly.

Linh Thi Xuan Phan, assistant professor in Computer and Information Science, has dedicated her research efforts to making the cyber-physical systems around us as functional, reliable and secure as possible.

“CPS often perform mission- or even life-critical functions, so a failure can have very serious consequences,” says Phan. “Using the example of airbags, cars have sensors that can detect a crash, and they have actuators that can inflate the airbags. It is really important for these systems to be safe and highly reliable. Inclusive of this is sensitivity to timing. In this example, it is not enough if the airbag inflates eventually after a crash. There is a very short window of time, measured in milliseconds, in which the actuator has to be triggered for the airbag to fully inflate by the time it comes in contact with the driver’s head.”

Because the consequences of a failure can be so severe, it is not enough to build CPS that work well in most cases; researchers need to be able to prove that a CPS has the required properties, even in the worst-case scenario. For instance, developers may need to prove that, after a crash, the controller in the car will always give the signal to inflate the airbag within a certain number of milliseconds, no matter what else it may have been doing when the crash occurred.

“Doing this kind of proof can be very difficult, especially for the complex systems we have today, and the system has to be built in a certain way to make a proof even possible,” says Phan. “It requires both sophisticated formal methods and special systems techniques to make this work, and these are the focus of my research.”

Not only is it imperative to design and implement CPS that are reliable in function and timing, researchers must also take into account that with increased connectivity comes security vulnerabilities. “There is a lot of concern now about CPS security,” says Phan. “Hackers are breaking into medical devices like insulin pumps, or power plants, and you may have seen a video from a recent research project at UCSD where someone remotely, using a wireless network, hacked into a car that was driving on the highway and took control of the power steering. We urgently need ways to make CPS devices more secure.”

Securing CPS and related devices (such as those that make up the emerging Internet of Things, or IoT) is very challenging. Traditional security techniques are often too expensive for CPS, which are typically small, embedded devices that have limited power, and there are attacks in the CPS space that the traditional techniques just can’t handle. “My group and I are working on stronger defenses that will make our infrastructure more secure and resilient,” says Phan. “This way, people won’t have to worry that a hacker will steer their car into a wall or knock out the local energy grid. In fact, we have an ongoing joint project with Intel in exactly that area.”

Phan came to Penn following the completion of her doctoral degree in Computer Science at the National University of Singapore, where she worked to accurately and efficiently model, analyze, and design complex real-time embedded systems. She joined Penn in 2009 as a research associate, eventually completing a postdoctoral fellowship in the in the Real-Time Systems group of the Penn Research in Embedded Computing and Integrated Systems Engineering (PRECISE) center. There, she continued the work she began during her doctoral studies and also focused on several new projects involving CPS and distributed systems.

“I am very excited to have Linh join the faculty in CIS,” says Sampath Kannan, Henry Salvatori Professor and chair of CIS. “Her research has been an excellent addition to the department and I look forward to seeing where she will take the field of Cyber-physical Systems in the coming years.”

Phan celebrates the collaborative atmosphere at Penn. “I love doing interdisciplinary work, and I think my research in particular can build bridges between several different areas,” she says. “I have a joint project with the Perelman School of Medicine on securing medical devices, and I am working with Christopher Yoo in Penn Law on legal and regulatory aspects of CPS.”

Within Penn Engineering, Phan’s group collaborates on CPS security and intrusion detection; on defending data centers against denial-of-service attacks; on diagnostics and forensics; and with members of the PRECISE center on real-time embedded systems. “The research environment is obviously fantastic,” she continues, “but above all, it is the people that made me decide to stay at Penn.”

Throughout her research, Phan has discovered that the knowledge and ideas related to improving CPS can also be applied to networks and data centers. Computing networks today, especially large ones such as those in data centers operated by Google and Facebook, do much more than ship data back and forth. They need to perform operations such as firewalling, intrusion detection, and data compression. Traditionally, these functions have been implemented via “middleboxes,” which are special-purpose devices that perform only a single function. “Think of these quite literally as lots of physical boxes in the data center that are connected to the rest of the network,” says Phan. “Middleboxes make the network more complicated to manage.”

Middleboxes are increasingly being replaced by virtual machines that run on a shared infrastructure, a bit like a modern cloud-computing platform such as Amazon Web Services. This is called network function virtualization, or NFV, and a few general-purpose computers can do all of the work. Phan is the principal investigator on a $1.1 million grant from the National Science Foundation for her proposal, “Network Functions Virtualization With Timing Guarantees,” in which she is collaborating with co-principal investigators Andreas Haeberlen and Boon Thau Loo, both associate professors in CIS at Penn. The project focuses on building a new real-time cloud infrastructure for NFV, which is a platform that can provide performance guarantees in a cloud computing setting. “My work on NFV and the real-time cloud should help make data centers work better,” says Phan. “The challenges are sufficiently similar to CPS in that we can hope to use some of the same ideas, but there are also important differences. The hope is that my group can find a way to adapt these ideas so that they can bring the same stability to NFV that they have already brought to CPS.”

In looking ahead, Phan states that it is important to take a long-term view. “I see security as a ‘grand challenge’ for my field,” she says. “There are a lot of vulnerable systems out there, and this is not something we can fix in 1-to-3 years because there are lots of hard research questions that we do not yet know how to answer. I think what we really need is to develop a way to fundamentally build these systems with rock-solid security. I also hope that my field will build more bridges to other areas. We need to rethink our approach and start taking a more interdisciplinary perspective.”