Cell is a distributed sorted in-memory store offering random access reads and writes, transactions, range queries, and more. It exploits RDMA for all read operations and standard message-passing for write operations to get the highest possible performance with modest system complexity. Cell expands on the lessons learned from the Pilaf project, which explored the properties of the system designs that will be prevalent in tomorrow’s datacenters. We believe that the interconnects in near-future datacenters will offer high bandwidth, very low latency, and Remote Direct Memory Access (RDMA).
Pilaf is an implementation of a high-performance distributed storage system over Infiniband. With existing Ethernet-based networking, many distributed system designs are impossible. Infiniband includes two key features that make new designs feasible: very low-latency, high-throughput messages, and RDMA reads and writes that do not involve the remote CPU. Pilaf is an in-memory distributed cache leveraging these key features. By comparing Pilaf to Memcached and other popular in-memory caches and key-value stores, we show the importance of considering new system designs to take advantage of networking hardware features rather than simply leveraging the higher throughput of such hardware for existing systems.
Applications of Convolutional Neural Networks to Facial Detection and Recognition in Wearable Computing and Augmented Reality
A real-time, fault-tolerant and easily deployable solution to detect rip currents using a wireless sensor network.
SimmsAI was a Markov Chain-based chatbot, capable of online training, research from web sources, and including a novel distributed work assignment and communication system
This NSF-funded academic research project created a physical, small-scale test environment to simulate wireless network scenarios.
In the summer of 2007, I participated in an REU (Research Experience for Undergraduates) at Stevens Institute of Technology in Hoboken, New Jersey. I developed a project that had previously been proposed by a graduate student at that institution but never attempted, using image processing methods to evaluate the probability that a photomicrography image contained cancerous cells. I spent ten weeks on the project, eventually producing a prototype in Matlab capable of identifying cell nuclei and thus predicting the presence of cancer cells in prostate photomicography even when faced with varied cell and image size, image contrast, image coloration, and other variables.