Zhuoran Wang

Implemented Dolev’s Reliable Communication algorithm and its five modifications described in the paper: Bonomi, G. Farina, and S. Tixeuil, “Multi-hop byzantine reliable broadcast with honest dealer made practical,” Journal of the Brazilian Computer Society, vol. 25, pp. 1–23, 2019
Implemented Bracha’s reliable broadcast algorithm on the top of the fully-connected network topology that Dolev algorithm provided.
Implemented a casual-order Byzantine broadcast algorithm using vector clock technique. causal-order Byzantine broadcast

Implemented Unix utilities in xv6, including sleep, find, xargs, gaining proficiency in system calls.
Added trace and Sysinfo system call
Optimized system calls and detected accessed pages.
Implemented Copy-on-Write Fork in xv6.
Developed an xv6 device driver for a network interface card (NIC).
Redesigned code to increase parallelism, addressing high lock contention in xv6 memory allocator.
Enhanced xv6 file system by adding support for large files and symbolic links.
Implemented mmap and munmap system calls in xv6.

Conducted profiling with gprof to identify bottlenecks; optimized matrix multiplication using SIMD (AVX) and multi-core parallelism (OpenMP), employing techniques including fine-tuning OMP directives and reordering multiplication to leverage cache locality.
Analyzed CPU and CUDA implementations, identified bottlenecks through Nsight system profiling, and optimized by improving data transfer efficiency, leveraging shared memory, and fine-tuning grid and block dimensions.
Accelerated the K Means algorithm using CUDA, implementing optimization strategies identified through profiling, including fine-tuning grid and block dimensions, and leveraging shared memory to reduce the data transfer overhead.

Approach 1 - Rayon Implementation: Developed a multi-threaded regular expression search tool in Rust. Used clap for command-line argument parsing, regex for pattern matching, and rayon for parallel processing. Ensured data consistency and thread safety in a multi threaded environment using Mutex and Arc.
Approach 2 - Thread Pool Implementation: Developed another multi-threaded regular expression search tool in Rust. Used clap for command-line argument parsing, regex for pattern matching, and implemented parallel processing using a custom thread pool. The custom thread pool employed mpsc for task scheduling and Mutex and Arc for thread safety and data consistency. Conducted performance testing on a Linux system using the time command. The custom thread pool approach achieved a 39.1% performance improvement over GNU grep when executing grep [mM]icrodevice ..

Investigated and optimized the classifier using a DPC-based data preprocessing method to address the problem of multi-classification unbalanced data.
Analyzed three mainstream classifiers: random forest, decision tree, k-nearest neighbor, and support vector machine, with a focus on using the weighted KNN algorithm based on kernel function for prediction training.
Utilized the DPC algorithm to select the cluster center and determine the optimal clipping threshold through experimentation, resulting in an optimized training set for the classifier.
Employed a cross test to test the accuracy of the model, and experimental results showed a well improvement in classifier accuracy using the optimization algorithm in most of cases.