| Modern computer systems play important roles in our society and everyday lives. Their performance, security and reliability are of critical importance. Real-world computer systems, however, occasionally suffer from performance degradation, security exploits, and poor reliability, because of the lack of efficient automatic analyses. This dissertation introduces a new methodology for building efficient automatic analyses for real-world computer systems through identifying and designing proper abstractions. It demonstrates the methodology within the context of three real-world computer systems: detecting net- work defects at the data plane level, exploiting data parallelism in web pages, and formally verifying security invariants in operating system kernels. This dissertation presents the design, implementation, and evaluation of the above systems, and shows that choosing the proper set of abstractions is an essential step to constructing efficient automatic analyses for real-world computer systems. Moreover, these analyses can become valuable tools to improve the performance, security and reliability of computer systems. |
