Power/Performance Bits: May 14

Detecting malware with power monitoring
Engineers at the University of Texas at Austin and North Carolina State University devised a way to detect malware in large-scale embedded computer systems by monitoring power usage and identifying unusual surges as a warning of potential infection.

The method relies on an external piece of hardware that can be plugged into the system to observe and monitor power usage. An external device is critical to avoid it being affected by an attack, say the researchers.

Certain power usage signatures can be identified as evidence of the presence of malware as well as determine how much of a threat they are to a compromised system.

“We know what power consumption looks like when embedded systems are operating at normal levels,” said Mohit Tiwari, an assistant professor at UT Austin. “By looking for power anomalies, we can tell with reasonable accuracy when malware is present in a system.”

White this approach should be able to detect many attacks, the researchers were also concerned about sophisticated malware that is designed to conceal their presence by mirroring the power usage of benign programs.

“The real technical contribution of this work has been our ability to successfully model malware that conceal themselves by mimicking the power signatures of benign programs,” Tiwari said. “Models of evasive malware can then be used to determine the extent of damage that power detectors can protect against.”

Tiwari also noted that while they can’t identify the specific kind of malware attacking a system, power anomalies can determine the level of threat.

“Malware keeps evolving in order to outsmart anti-virus software, meaning engineers must also continuously retrain their programs,” said Shijia Wei, a Ph.D. candidate in the department of electrical and computer engineering at UT Austin. “With our device, we can force the malware to mimic benign programs on embedded systems, and this can greatly reduce the potential damage an attack can cause.”

“We found that the effort required to mimic normal power consumption and evade detection forced malware to slow down its data transfer rate by between 86 and 97 percent,” said Aydin Aysu, an assistant professor of electrical and computer engineering at NC State. “In short, our approach can still reduce the effects of malware, even in those few instances where the malware is not detected.”

Colorful, tunable LEDs
Researchers from Lehigh University, West Chester University, Osaka University, and University of Amsterdam created gallium nitride (GaN)-based LEDs with simple color tuning by changing the time sequence at which the operation current is provided to the device.

The technique is compatible with current LEDs that are at the core of commercial solid state LED lighting.

Rather than using three or four individual LEDs close together, the full color spectrum could be achieved with just a single LED, said Volkmar Dierolf, distinguished professor and chair of the Department of Physics at Lehigh. “We show that is possible to attain red, green and blue emissions originating from just one GaN LED-structure that uses doping with a single type of rare earth ion, Europium (Eu). Using intentional co-doping and energy-transfer engineering, we show that all three primary colors can emit due to emission originating from two different excited states of the same Eu3+ ion (~620 nm and ~545nm) mixed with near band edge emission from GaN centered at ~430nm. The intensity ratios of these transitions can be controlled by choosing the current injection conditions such as injection current density and duty cycle under pulsed current injection.”

Top row: A GaN:Eu LED, which can be tuned from red-yellow due to red and green light mixing from different Eu states. Middle and bottom rows: A GaN:Eu LED with additionally added Si/Mg, which adds blue emission. Each picture is under a different current injection/filtering condition. (Image courtesy of West Chester University, Lehigh University)

“The main idea of this work—the simultaneous active exploitation of multiple excited states of the same dopant—is not limited to the GaN:Eu system, but is more general,” added Brandon Mitchell, an assistant professor in the Department of Physics and Engineering at West Chester University. “The presented results could open up a whole new field of tunable emission of colors from a single dopant in semiconductors, which can be reached by simple injection current tuning.”

Beyond color, the technique would allow commercial LEDs to be switched between bright white and warm white light. Dierolf noted that “it would also be beneficial for micro-LED displays, since it allows for higher density of pixels.”

The technique is compatible with current commercial GAN-based LEDs.

Supercomputer load balancing
Computer scientists at Virginia Tech applied machine learning to the problem of load balancing data processing tasks across the thousands of servers that make up a supercomputer. By incorporating machine learning to predict not only tasks but types of tasks, researchers found that load on various servers can be kept balanced throughout the entire system.

Typically, supercomputer data management systems rely on approaches that assign tasks in a round-robin manner to servers without regard to the kind of task or amount of data, which can result in degraded performance while the system waits for slower tasks.

Instead, the team’s approach uses an end-to-end control plane that combined the application-centric strengths of client-side approaches with the system-centric strengths of server-side approaches.

This approach allowed the team to monitor the system and allowed the data storage system to learn and predict when larger loads might be coming down the pike or when the load became too great for one server. The system also provided real-time information in an application-agnostic way, creating a global view of what was happening in the system.

“This study was a giant leap in managing supercomputing systems. What we’ve done has given supercomputing a performance boost and proven these systems can be managed smartly in a cost-effective way through machine learning,” said Bharti Wadhwa, a Ph.D. candidate in the Department of Computer Science at Virginia Tech. “We have given users the capability of designing systems without incurring a lot of cost.”

Importantly, the end-to-end system allowed users to benefit from the load balanced setup without changing the source code.

The end-to-end control plane consisted of storage servers posting their usage information to the metadata server. An autoregressive integrated moving average time series model was used to predict future requests with approximately 99 percent accuracy and were sent to the metadata server in order to map to storage servers using minimum-cost maximum-flow graph algorithm.

“The algorithm predicted the future requests of applications via a time-series model,” said Arnab K. Paul, a Ph.D. candidate in the Department of Computer Science at Virginia Tech. “This ability to learn from data gave us a unique opportunity to see how we could place future requests in a load balanced manner.”