Defense Notices

The mobile devices emerge as the most popular computing platform since 2011. Different from the traditional PC, the mobile devices are more power-constraint and performance-sensitive due to its size. In order to reduce the power consumption and improve the performance, we focus on the Last Level Cache (LLC), which is the power-hungry structure and critical to the performance in mobile platforms. In this project, we first integrate the McPAT power model into the Gem5 simulator. We also introduce the emerging memory technologies, such as Sprin-Transfer Torque RAM (STT-RAM) and embedded DRAM (eDRAM), into the cache design and compare their power and performance effectiveness with the conventional SRAM-based cache. Additionally, we identify that the frequent execution switch between the kernel and user code is the major reason for the high LLC miss in mobile applications. This is because blocks belonging to kernel and user space have severe interferences. We further propose static and dynamic way partition schemes to separate the cache blocks from kernel and user space. The experiment results show promising power reduction and performance improvement with our proposed techniques.

Recent years have witnessed the widespread adoption of managed programming languages that are designed to execute on virtual machines. Virtual machine architectures provide several powerful software engineering advantages over statically compiled binaries, such as portable program represenations, additional safety guarantees, and automatic memory and thread management, which have largely driven their success. To support and facilitate the use of these features, virtual machines implement a number of services that adaptively manage and optimize application behavior during execution. Such runtime services often require tradeoffs between efficiency and effectiveness, and different policies can have major implications on the system's performance and energy requirements. 

In this work, we extensively explore policies for the two runtime services that are most important for achieving performance and energy efficiency: dynamic (or Just-In-Time (JIT)) compilation and memory management. First, we examine the properties of single-tier and multi-tier JIT compilation policies in order to find strategies that realize the best program performance for existing and future machines. We perform hundreds of experiments with different compiler aggressiveness and optimization levels to evaluate the performance impact of varying if and when methods are compiled. Next, we investigate the issue of how to optimize program regions to maximize performance in JIT compilation environments. For this study, we conduct a thorough analysis of the behavior of optimization phases in our dynamic compiler, and construct a custom experimental framework to determine the performance limits of phase selection during dynamic compilation. Lastly, we explore innovative memory management strategies to improve energy efficiency in the memory subsystem. We propose and develop a novel cross-layer approach to memory management that integrates information and analysis in the VM with fine-grained management of memory resources in the operating system. Using custom as well as standard benchmark workloads, we perform detailed evaluation that demonstrates the energy-saving potential of our approach.

GPGPUs (general-purpose computing on graphics processing units) emerge as a highly attractive platform for HPC (high performance computing) applications due to its strong computing power. Unlike the graphic processing applications, HPC applications have rigorous requirement on execution correctness, which is generally ignored in the traditional GPU design. Soft Errors, which are failures caused by high-energy neutron or alpha particle strikes in integrated circuits, become a major reliability concern due to the shrinking of feature sizes and growing integration density. In this project, we first build a framework GPGPU-SODA to model the soft-error vulnerability of GPGPU microarchitecture using a publicly available simulator. Based on the framework, we identified the streaming processors are reliability hot-spot in GPGPUs. We further observe that the streaming processors are not fully utilized during the branch divergence and pipeline stalls caused by the long latency operations. We then propose a technique RISE to recycle the streaming processors idle time for soft-error detection in GPGPUs. Experimental results show that RISE obtains the good fault coverage with negligible performance degradation.

Digital Audio Workstation (DAW) applications are real-time applications that have special timing constraints. HGS is a real-time scheduling framework that allows developers implement custom schedulers based on any scheduling algorithm through a process of direct interaction between client threads and their schedulers. Such scheduling could extend well beyond the common priority model that currently exists and could be a representation of arbitrary application semantics that can be well understood and acted upon by its associated scheduler. We like to term it "need based scheduling". In this thesis we firstly study some DAW implementations and later create a few different HGS schedulers aimed at assisting DAW applications meet their needs.

Real-time, precise range measurement between objects is useful for a variety of applications. The slow propagation of acoustic signals (330 m/s) in air makes the use of ultrasound frequencies an ideal approach to measure an accurate time of flight. The time of flight can then be used to calculate the range between two objects. The objective of this project is to achieve a precise range measurement within 10 cm uncertainty and an update rate of 30 ms for distances up to 10 m between unmanned aerial vehicles (UAVs) when flying in formation. Both transmitter and receiver are synchronized with a 1 pulse per second signal coming from a GPS. The time of flight is calculated using the cross-correlation of the transmitted and received waves. To allow for various users, a 40 kHz signal is phase modulated with Gold or Kasami codes.

Ice shelves are sensitive indicators of climate change and play a critical role in the stability of ice sheets and oceanic currents. Basal melting of ice shelves affect both the mass balance of the ice sheet and the global climate system. This melting and refreezing influences the development of Antarctic Bottom Water, which help drive the oceanic thermohaline circulation, a critical component of the global climate system. Basal melt rates can be estimated through traditional glaciological techniques, relying on conversation of mass. However, this requires accurate knowledge of the ice movement, surface accumulation and ablation, and firn compression. Boreholes can provide direct measurement of melt rates, but only provide point estimates and are difficult and expensive to perform. Satellite altimetry measurements have been heavily relied upon for the past few decades. Thickness and melt rate estimates require the same conservation of mass a priori knowledge, with the additional assumption that the ice shelf is in hydrostatic equilibrium. Even with newly available, ground truthed density and geoid estimates, satellite data derived ice shelf thickness and melt rate estimates suffers from relatively course spatial resolution and interpolation induced error. Non destructive radio echo sounding (RES) measurements from long range airborne platforms provide best solution for fine spatial and temporal resolution over long survey traverses and only require a priori knowledge of firn density and surface accumulation. Previously, RES data derived basal melt rate experiments have been limited to ground based experiments with poor coverage and spatial resolution. To improve upon this, an airborne multi channel wideband radar has been developed for the purpose of imaging shallow ice and ice shelves. A moving platform and cross track antenna array will allow for fine resolution 3 D imaging of basal topography. An initial experiment will use a ground based system to image shallow ice and generate 3 D imagery as a proof of concept. This will then be applied to ice shelf data collected by an airborne system.

Residing between the network layer and the application layer, the transport 
layer exchanges application data using the services provided by the network. Given the unreliable nature of the underlying network, reliable data transfer has become one of the key requirements for those transport-layer protocols such as TCP. Studying the various mechanisms developed for TCP to increase the correctness of data transmission while fully utilizing the network's bandwidth provides us a strong background for our study and development of our own resilient end-to-end transport protocol. Given this motivation, in this thesis, we study the dierent 
TCP's error control and congestion control techniques by simulating them under dierent network scenarios using ns-3. For error control, we narrow our research to acknowledgement methods such as cumulative ACK - the traditional TCP's way of ACKing, SACK, NAK, and SNACK. The congestion control analysis covers some TCP variants including Tahoe, Reno, NewReno, Vegas, Westwood, Westwood+, and TCP SACK.

Mobile traffic is expected to grow at an annual compound rate of 66% in the next 3 years, while among the data types that account for this growth mobile video has the highest growth rate. Since most video applications are delay-sensitive, the delay-sensitive traffic will be the dominant traffic over future wireless communications. Consequently, future mobile wireless systems will face the dual challenge of supporting large traffic volume while providing reliable service for various kinds of delay-sensitive applications (e.g. real-time video, online gaming, and voice-over-IP (VoIP)). Past work on delay-sensitive communications has generally overlooked the physical-layer considerations such as modulation and coding scheme (MCS), probability of decoding error, and coding delay by employing oversimplified models for the physical-layer. With the proposed research we aim to bridge information theory, communication theory, and queueing theory by jointly considering the delay-violation probability and the probability of decoding error to identify the fundamental trade-offs among wireless system parameters such as channel fading speed, average received signal-to-noise ratio (SNR), MCS, and user perceived quality of service. We will model the underlying wireless channel by a finite-state Markov chain, use channnel dispersion to track the probability of decoding error and the coding delay for a given MCS, and focus on the asymptotic decay rate of buffer occupancy for queueing delay analysis. The proposed work will be used to obtain fundamental bounds on the performance of queued systems over wireless communication channels.

The high level of functionality and flexibility of modern wideband wireless networks, LTE and WiMAX, have made them the preferred technology for providing mobile internet connectivity. The high performance of these systems comes from adopting several innovative techniques such as Orthogonal Frequency Division Multiplexing (OFDM), Automatic Modulation and Coding (AMC), and Hybrid Automatic Repeat Request (HARQ). However, this flexibility also opens the door for network exploitation by other ad-hoc networks, like Device-to-Device technology, or covert systems. In this work effort, we provide the theoretical foundation for a new ad-hoc wireless covert system that hides its transmission in the RF spectrum of an OFDM-based wideband network (Target Network), like LTE. The first part of this effort will focus on designing the covert waveform to achieve a low probability of detection (LPD). Next, we compare the performance of several available detection methods in detecting the covert transmission, and propose a detection algorithm that would represent a worst case scenario for the covert system. Finally, we optimize the performance of the covert system in terms of its throughput, transmission power, and interference on/from the target network.

Computer networks are prone to targeted attacks and natural disasters that could disrupt its normal operation and services. Adding links to form a full mesh yields the most resilient network but it incurs unfeasible high cost. In this research, we investigate the resilience improvement of real-world network via adding a cost-efficient set of links. Adding a set of link to get optimal solution using exhaustive search is impracticable given the size of communication network graphs. Using a greedy algorithm, a feasible solution is obtained by adding a set of links to improve network connectivity by increasing a graph robustness metric such as algebraic connectivity or total path diversity. We use a graph metric called flow robustness as a measure for network resilience. To evaluate the improved networks, we apply three centrality-based attacks and study their resilience. The flow robustness results of the attacks show that the improved networks are more resilient than the non-improved networks.