Thesis: B.S., Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science, 1980 / by Hong Kien Ong. / B.S. / B.S. Massachusetts Institute of Technology, Department of Electrical Engineering and Computer Science

We present the Hadley model, a formal descriptive model of input and output for modern computer operating systems. Our model is intentionally inspired by the Open Systems Interconnection model of networking; I/O as a process is defined as a set of translations between a set of computer-sensible forms, or layers, of information. To illustrate an initial application domain, we discuss the utility of the Hadley model and a potential associated I/O system as a tool for digital forensic investigators. To illustrate practical uses of the Hadley model we present the Hadley Specification Language, an essentially functional language designed to allow the translations that comprise I/O to be written in a concise format allowing for relatively easy verifiability. To further illustrate the utility of the language we present a read/write Microsoft DOS FAT12 and read-only Linux ext2 file system specification written in the new format. We prove the correctness of the read-only side of these descriptions. We present test results from operation of our HSL-driven system both in user mode on stored disk images and as part of a Linux kernel module allowing file systems to be read. We conclude by discussing future directions for the research.

Control-flow hijack attacks on software exploit vulnerabilities in the software’s memory handling. Over the years, various security mitigations have been developed to counter these attacks. However, compatibility issues have hindered the adoption of such measures in some legacy systems. This thesis focuses on the case of the legacy DOS system and examines whether a DOS system running the DOS/4GW protected mode extender can provide control-flow protection against an attack exploiting a buffer overflow vulnerability in the well-known retro game Duke Nukem3D. To investigate this, three model programs were created, and designed with memory models that share memory layout characteristics with the target retro game’s executable. Experimental attacks were then conducted on these models, aiming to identify an effective attack vector for the target vulnerability. The underlying theory suggests that memory models that segregate application data into distinct memory segments could potentially safeguard against the demonstrated attack. However, attempts to implement such a memory model within an application proved unsuccessful. The challenge that remains is to prove the existence of memory models under DOSprotected mode that can effectively shield Duke Nukem 3D, or other legacy games, from the control-flow hijack attack demonstrated in this thesis.

The increasing demand for storage and computation has driven the growth of large data centers--the massive server farms that run many of today's Internet and business applications. A data center can comprise many thousands of servers and can use as much energy as a small city. The massive amounts of computation power contained in these systems results in many interesting distributed systems and resource management problems. In this thesis we investigate challenges related to data centers, with a particular emphasis on how new virtualization technologies can be used to simplify deployment, improve resource efficiency, and reduce the cost of reliability, all in application agnostic ways. We first study problems that relate to the initial capacity planning required when deploying applications into a virtualized data center. We demonstrate how models of virtualization overheads can be utilized to accurately predict the resource needs of virtualized applications, allowing them to be smoothly transitioned into a data center. We next study how memory similarity can be used to guide placement when adding virtual machines to a data center, and demonstrate how memory sharing can be exploited to reduce the memory footprints of virtual machines. This allows for better server consolidation, reducing hardware and energy costs within the data center. We then discuss how virtualization can be used to improve the performance and efficiency of data centers through the use of "live'' migration and dynamic resource allocation. We present automated, dynamic provisioning schemes that can effectively respond to the rapid fluctuations of Internet workloads without hurting application performance. We then extend these migration tools to support seamlessly moving applications across low bandwidth Internet links. Finally, we discuss the reliability challenges faced by data centers and present a new replication technique that allows cloud computing platforms to offer high performance, no data loss disaster recovery services despite high network latencies.

This dissertation is motivated by an intersection of changes occurring in modern software and hardware; driven by increasing application performance and energy requirements while Moore's Law and Dennard Scaling are facing challenges of diminishing returns. To address these challenging requirements, new features are increasingly being packed into hardware to support new offloading capabilities, as well as more complex software policies to manage these features. This is leading to an exponential explosion in the number of possible configurations of both software and hardware to meet these requirements. For network-based applications, this thesis demonstrates how these complexities can be tamed by identifying and exploiting the characteristics of the underlying system through a rigorous and novel experimental study. This thesis demonstrates how one can simplify this control strategy problem in practical settings by cutting across the complexity through the use of mechanisms that exploit two fundamental properties of network processing. Using the common request-response network processing model, this thesis finds that controlling 1) the speed of network interrupts and 2) the speed at which the request is then executed, enables the characterization of the software and hardware in a stable and well-structured manner. Specifically, a network device's interrupt delay feature is used to control the rate of incoming and outgoing network requests and a processor's frequency setting was used to control the speed of instruction execution. This experimental study, conducted using 340 unique combinations of the two mechanisms, across 2 OSes and 4 applications, finds that optimizing these settings in an application-specific way can result in characteristic performance improvements over 2X while improving energy efficiency by over 2X.

No description available.

No description available.

Most operating systems provide protection and isolation to user processes, but not to critical system components such as device drivers or other systems code. Consequently, failures in these components often lead to system failures. VirtuOS is an operating system that exploits a new method of decomposition to protect against such failures. VirtuOS exploits virtualization to isolate and protect vertical slices of existing OS kernels in separate service domains. Each service domain represents a partition of an existing kernel, which implements a subset of that kernel's functionality. Service domains directly service system calls from user processes. VirtuOS exploits an exceptionless model, avoiding the cost of a system call trap in many cases. We illustrate how to apply exceptionless system calls across virtualized domains. To demonstrate the viability of VirtuOS's approach, we implemented a prototype based on the Linux kernel and Xen hypervisor. We created and evaluated a network and a storage service domain. Our prototype retains compatibility with existing applications, can survive the failure of individual service domains while outperforming alternative approaches such as isolated driver domains and even exceeding the performance of native Linux for some multithreaded workloads. The evaluation of VirtuOS revealed costs due to decomposition, memory management, and communication, which necessitated a fine-grained analysis to understand their impact on the system's performance. The interaction of virtual machines with multiple underlying software and hardware layers in virtualized environment makes this task difficult. Moreover, performance analysis tools commonly used in native environments were not available in virtualized environments. Our work addresses this problem to enable an in-depth performance analysis of VirtuOS. Our Perfctr-Xen framework provides capabilities for per-thread analysis with both accumulative event counts and interrupt-driven event sampling. Perfctr-Xen is a flexible and generic tool, supports different modes of virtualization, and can be used for many applications outside of VirtuOS. / Ph. D.

Multitasking systems are becoming increasingly used for implementing real-time systems since they are well-suited to asynchronous, often overlapping, events. With the availability of kernels such as Hunter and Ready's VRTX Operating System components, multitasking becomes a good alternative to other implementations of real-time systems, such as interrupt-driven and polling systems. We developed a software design method and wrote a sample real-time system with many of the characteristics of typical real-time systems. This multitasking system uses a standard printer port to output bits which indicate the various activities of the tasks in the system. It also allows the user to interactively change the priorities of the tasks and to modify parameters which determine how long and how often the tasks execute. Through the use of the printer port connected to a logic analyzer and the ability to change various parameters in the system dynamically, the interaction between tasks was studied for different situations. The observed interaction between tasks was consistent with intuition. We observed task pre-emption, tasks waiting for results from other tasks and the system overhead required for context switching. When the timing restrictions were increased, we observed that data is lost during intertask communication and that higher priority tasks are the only tasks which run. This ability to observe the interaction between tasks has removed much of the mystery surrounding multitasking. / M.S.

Resource Management is an extremely important concern for the network manager. Of the bus, ring and the star network topologies, usually employed in a local area network, the bus topology provides the highest reliability. However, peer-to-peer protocols usually followed in such a topology, make it extremely difficult to incorporate resource management features at a lower level of network architecture. This thesis presents design considerations for a session-level network controller, to be implemented on a local area network with a bus topology. The controller, will provide essential resource management, and attempt to improve the throughput of the network. The design is based on a single-board computer, and a streamlined operating system is also included in the design. It is claimed that such an environment will be ideal for networking tasks, like the controller presented. The design is software-oriented, though device selection is discussed, both for the controller and its network interface. The local area network used for the development of this project is a sixteen hundred node campus network at Virginia Tech ( using LocalNet 20) supplied by SYTEK, Inc. / M.S.