Strategies for divisible loads that are computationally intensive
The main research topic of my PhD work was in the design of scheduling algorithms in parallel and distributed systems. The work mainly focused on the design of scheduling strategies, using Divisible Load Theory (DLT), for divisible loads or tasks that are computationally intensive. I considered the design and analysis of load distribution strategy in generic scenarios but later, I expanded the work to specific scheduling problems in the field of bioinformatics.
Our design simplifies the management of heterogeneous storage systems
After I graduated, I joined the Network Storage Technology (NST) lab of the Data Storage Institute (DSI), part of the Singapore Agency for Science, Technology and Research (A*STAR) (key agency research development for the Singapore government). I was involved in the pilot project of design, research and development of a dynamic Grid storage resource allocation infrastructure. In our design, we considered the Storage Management Initiative Specifications (SMI-S) that simplify the management of heterogeneous storage systems/devices, such that devices can be added, replace, or migrate from one server to another with minimal or no reconfiguration at all.
Later, I joined the University of Exeter, UK, on project with the University of Manchester. The project, referred to as e-fungi, has the primary aim of developing a data warehouse that integrates sequence and functional data from multiple fungal sequences in a way that facilitates the systematic study and analysis of species with reference to model organisms that have been more fully explored in terms of functional characteristics. More information can be found on the e-fungi official page.
Presented the characteristics of nucleosome by means of evolutions
In 2008, I went to the University of Cambridge to research quadruplex sequences in the human genome. Quadruplex sequences, also known as G-quadruplex, are sequences that are able to form the 'interesting' G-tetrad structure. These sequences, prevalent in promoter regions, have been found to have gene regulating functions. Further, due to their unique structure, quadruplex sequences are especially good candidates for drug targets. During those years, I have identify list of quadruplex targets for cancer related genes, showed the associations of quadruplex with nucleosomes, presented the characteristics of nucleosome by means of evolutions, etc. I have also extebded various tools for the web-based quadruplex research community. These tools and other details can be found in Quadruplex.org
Mathematics and technologies in the financial world.
I am currently working in G0ldman Sachs. In the first year, I worked on the Foreign Exchange (FX) Straight Through Procossing (STP) system. STP is part of the effort of making electronic trading (e-trading) to be executed at high frequency. While STP often view as just a reconciliation process for trades, it plays an important part in risk management. In the project, we looked into the current design and implementation of the STP process, and extends the process into support multiple protocols. We also introduced multi-threading capability into the process to achieve lower delays and higher throughput.
In 2012, I worked in developing a replacement market making system for our FX business as part of a strategic movement to replace our aging infrastructure. The challenge of the project lies on the importance, age, and complexity of our market making system. To put into perspective, the old system was developed in around 8 years ago. During the time when the system is being used, new products and new logics are constantly being added into the systems. Each products and logics can be written by a totally different team with different coding styles, and new logics added everywhere, often in a place that you least expected. To make the problem more challenging, many of the developers of the old systems, already left the company.
Beautifully engage both price and time
Whilst being a challenging project, this is also a project that I can find usage of my algorithm design skill. One of the new feature that we added to the system is the 'hysteresis' logic which controls how prices we quoted gets updated. The challenge is to come up with an algorithm gives the most prices updates (ie selective price update) without exceeding the limit of the recipients whilst ensuring that the price quoted are fair. I have ran through a number of algorithms that can be used, and analyze the merits for each of them. Whilst time is frequently the element to be considered in hysteresis algorithm, the final algorithm we used, beautifully engage both the price and time.