Author's School

School of Engineering & Applied Science

Author's Department/Program

Electrical and Systems Engineering

Language

English (en)

Date of Award

5-24-2012

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Chair and Committee

Paul Min

Abstract

We propose solutions for application virtualization to mitigate the performance loss in streaming and browser-based applications. For the application streaming, we propose a solution which keeps operating system components and application software at the server and streams them to the client side for execution. This architecture minimizes the components managed at the clients and improves the platform-level incompatibility. The runtime performance of application streaming is significantly reduced when the required code is not properly available on the client side. To mitigate this issue and boost the runtime performance, we propose prefetching, i.e., speculatively delivering code blocks to the clients in advance. The probability model on which our prefetch method is based may be very large. To manage such a probability model and the associated hardware resources, we perform an information gain analysis. We draw two lower bounds of the information gain brought by an attribute set required to achieve a prefetch hit rate. We organize the probability model as a look-up table: LUT). Similar to the memory hierarchy which is widely used in the computing field, we separate the single LUT into two-level, hierarchical LUTs. To separate the entries without sorting all entries, we propose an entropy-based fast LUT separation algorithm which utilizes the entropy as an indicator. Since the domain of the attribute can be much larger than the addressable space of a virtual memory system, we need an efficient way to allocate each LUT's entry in a limited memory address space. Instead of using expensive CAM, we use a hash function to convert the attribute values into addresses. We propose an improved version of the Pearson hashing to reduce the collision rate with little extra complexity. Long interactive delays due to network delays are a significant drawback for the browser-based application virtualization. To address this, we propose a distributed infrastructure arrangement for browser-based application virtualization which reduces the average communication distance among servers and clients. We investigate a hand-off protocol to deal with the user mobility in the browser-based application virtualization. Analyses and simulations for information-based prefetching and for mobile applications are provided to quantify the benefits of the proposed solutions.

Comments

Permanent URL: http://dx.doi.org/10.7936/K7BZ6438

Download

Share

COinS