Large-scale network simulation has grown in importance due to a rapid increase in Internet size and the availability of Internet measurement topologies with applications to computer networks and network security. A key obstacle to large-scale network simulation over PC clusters is the memory balancing problem, where a memory-overloaded machine can slow down a distributed simulation due to disk I/O overhead. Network partitioning methods for parallel and distributed simulation are insufficiently equipped to handle new challenges brought on by memory balancing due to their focus on CPU and communication balancing.

This dissertation studies memory balancing for large-scale network simulation in power-law networks over PC clusters. First, we design and implement a measurement subsystem for dynamically tracking memory consumption in DaSSFNet, a distributed network simulator. Accurate monitoring of memory consumption is difficult due to complex protocol interaction through which message related events are created and destroyed inside and outside a simulation kernel. Second, we achieve efficient memory cost monitoring by tackling the problem of estimating peak memory consumption of a group of simulated network nodes in power-law topologies during network partitioning. In contrast to CPU balancing where the processing cost of a group of nodes is proportional to their sum, in memory balancing this closure property need not hold. Power-law connectivity injects additional complications due to skews in resource consumption across network nodes. Third, we show that the maximum memory cost metric outperforms the total cost metric for memory balancing under multilevel recursive partitioning but the opposite holds for CPU balancing. We show that the trade-off can be overcome through joint memory-CPU balancing—-in general not feasible due to constraint conflicts—-which is enabled by network simulation having a tendency to induce correlation between memory and CPU costs. Fourth, we evaluate memory balancing in the presence of virtual memory (VM) management which admits larger problem instances to be run over limited physical memory. VM introduces complex memory management dependencies that make understanding and evaluating simulation performance difficult. We provide a performance evaluation framework wherein the impact of memory thrashing in distributed network simulation is incorporated which admits quantitative performance comparison and diagnosis. Fifth, we show that improved memory balancing under the maximum cost metric in the presence of VM manifests as faster completion time compared to the total cost metric despite the CPU balancing advantage of the latter. In the cases where the CPU balancing advantage of the total cost metric is strong, we show that joint memory-CPU balancing can achieve the best of both worlds.

We carry out performance evaluation using benchmark applications with varying traffic characteristics: BGP routing, worm propagation under local and global scanning, and distributed client/server system. We use a testbed of 32 Intel x86 machines running a measurement-enhanced DaSSFNet over Linux.

American society in the present day is grappling with issues of privacy and surveillance. These issues, the technologies involved, and implications for the organization and function of American society are examined in this dissertation.

Public Law 109-13 contains the Real ID Act, and the implementation of this act has far-reaching ramifications for Americans’ privacy. The Real ID Act, an exemplar of recentlaws regarding privacy and surveillance, serves as a basis for discussing the development of a surveillance society and its potential harm to American citizens.

The dissertation begins by framing the evolution of the concept in American society, exploring anti-terror legislation as the latest assumption of extraordinary powers by the state in times of war and national emergency, and comparing previous abridgements of enumerated Constitutional rights in such times.

It next discusses the implication of the Real ID as an insecure collection of databases, and then it examines the effect of Real ID on American citizens’ privacy as a national identification card. States have resisted the implementation of the act on the bases that the act constitutes an unfunded mandate and damages privacy.

The new surveillance system erodes personal privacy and creates a threat to privacy and autonomy from both criminals and the government, or sometimes (due to insider abuse of data) both. The dissertation details the possibility of how Real ID information access can be used against people in ways both legal and illegal, with comparisons to Great Britain; it also questions whether the government is even capable of handling increased information resources or whether such resources only provide more opportunities for improper access and misuse of personal data.

For most people, the developing surveillance state may only pose potential danger until someone is identified as a target, but its potential chilling effect threatens participatory democracy and the expression of legitimate political dissent.

The goal of this dissertation is to increase awareness of the incremental erosion of privacy rights which, once surrendered, become increasingly difficult to regain. It also aims to question some of the security assumptions that justify this erosion.

Existing work on privacy-preserving data publishing cannot satisfactorily prevent an adversary with background knowledge from learning important sensitive information. The main challenge lies in modeling the adversary’s background knowledge. We propose a novel approach to deal with such attacks. In this approach, one first mines knowledge from the data to be released and then uses the mining results as the background knowledge when anonymizing the data. The rationale of our approach is that if certain facts or background knowledge exist, they should manifest themselves in the data and we should be able to find them using data mining techniques. One intriguing aspect of our approach is that one can argue that it improves both privacy and utility at the same time, as it both protects against background knowledge attacks and better preserves the features in the data. We then present the Injector framework for data anonymization. Injector mines negative association rules from the data to be released and uses them in the anonymization process. We also develop an efficient anonymization algorithm to compute the injected tables that incorporates background knowledge. Experimental results show that Injector reduces privacy risks against background knowledge attacks while improving data utility.