PhD candidate Dongyuan Zhan talks about his research and shares his experience as a graduate student in the Department of Computer Science. The following includes information about his research on multicore processors, advice for students considering graduate school, and the impact of faculty advisors.
Bits & Bytes (BB): Can you explain your current research/dissertation topic?
Dongyuan Zhan (DZ): My current research/dissertation topic is “Holistically Cooperative Resource Management In Many-Core Systems” [1]. The idea of “Holistic Cooperative Resource Management” is in relation to “Autonomous Resource Management”. Here is a little background: a computer system can typically be viewed as a hierarchy system with such modular layers as applications, runtime systems (e.g., virtual machines), operating systems and architecture (e.g., various hardware components). An autonomous approach is targeted at managing resources transparently at an individual layer without knowing other layers’ management decisions. For instance, autonomous cache management, a hot topic in the computer architecture research community, focuses on profiling cache access streams and making cache management decisions accordingly, both at the hardware level. It has no knowledge of running applications’ semantic characteristics that are available at the software level. This kind of approach favors the modularity of a layer, but my motivational experiments have shown that it only provides suboptimal performance due to the lack of cross-layer cooperation. My dissertation research is attempting to enabling different layers to cooperate with each other in managing resources and thus providing globally optimal performance.
BB: Can you tell us about the importance of the resource management of multicore processors and what sparks your interest in them?
DZ: Multicore systems are almost already everywhere. For instance, most personal computers, mobile phones and tablets have dual or quad cores on a single processor chip. As the transistor size keeps shrinking, we can expect more and more cores to be integrated on a processor die in the near future. In other words, we are advancing towards a many-core era. An example is the emerging commodity 100-core processor TILE-Gx100 by Tilera Corporation [2]. But the ever-increasing core count on a chip does not necessarily mean scalable performance mainly due to severer resource contention. This is because the core count is exponentially scalable while the number of resources (e.g., cache capacity and interconnect bandwidth) required by the cores is mostly linearly scalable. Without judicious management of resources for many-core processors, those cores will adversely compete with each other for cache capacity, interconnect bandwidth, memory bandwidth, etc., which has a detrimental influence on the scalability of many-core systems’ overall performance.
My enthusiasm for many-core resource management is because of its significant performance impact. I would like to make my own contributions to better resource management solutions for many-core systems.
BB: What is your research approach like? Can you explain the process of writing a dissertation?
DZ: In a word, my research approach is almost always “top-down”. This is a result of the nature of system research. More specifically, I need to first establish a big picture of the entire system, know what kind of components constitutes the system and how they are related to each other, and find out where the performance bottleneck is rooted. This is a process of problem identification, which also involves doing literature review and gathering quantitative evidence. After demonstrating how significant the problem is and analyzing why previous approaches do not work (well), we will try to propose our own potential solutions, which is a process of problem solving. Each of the potential solutions needs to be evaluated with real applications and compared among themselves and against prior-arts. The evaluation has to be thorough so as to be insightful and convincing. Most of the time, there should be feedback from solution evaluation to problem solving, because a solution may need to be improved according to the evaluation results. This should continue until satisfactory solutions that meet design objectives are found.
In my mind, writing a dissertation thesis is essentially to shed light on the fundamentals and essentials of our research by demonstrating steady and smooth progression from problem identification to problem solving, and then to solution evaluation. The organization of a dissertation thesis should also follow the “top-down” style.
BB: Your advisors are Dr. Hong Jiang and Dr. Sharad Seth. What influence have they had on you both professionally and personally?
DZ: Dr. Hong Jiang and Dr. Sharad Seth are both great tutors. Professionally, I learned a lot of research methodology, skills and knowledge from them. Dr. Hong Jiang is very sharp-minded. He taught me a lot about discovering a research niche, logic, arguments and reasoning. Dr. Seth has a very solid mathematical background. From him, I learned how rigorous research progression can be made. Personally, they are good people with great personalities, and I gradually learned from them how to live a harmonious and balanced life with research being the career focus.
BB: What has been your overall experience in the UNL CSE graduate program?
DZ: My feeling about our graduate program has always been great. We have experienced and knowledgeable faculty members, from whom I can learn about the cyber world by attending their classes and through discussions with them. The department staff is quite helpful. They provide us considerate support with system administration, student affairs and various activities and events. The research facilities here are also great. We have powerful supercomputers in the Holland Computing Center [3] such as “Tusker” (with 6784 processor cores), “Firefly” (with 5600 processor cores) and “Sandhills” (with 1440 processor cores).
BB: What advice would you give undergraduates who are considering applying to graduate school?
DZ: For the undergraduates who are interested in graduate problems, here are my three pieces of advice. First, you will need to have a rough idea of what your future career goal is, what your interest is in and whether attending a graduate program can help fulfill your goal and interest. In other words, you will need to be strongly motivated for graduate study, since it can be very challenging and may take a lot of time and energy. Second, you’d better decide which professor you want to work with. Do some research on his/her funding, publications, projects and group, talk with him/her and other group members about what you and the professor expect from each other, and find out what the group culture is like. Third, be flexible and prepared to adjust yourself to the new environment once you are enrolled.
BB: What qualities do you consider most important in order to be successful in academia?
DZ: Being optimistic, confident and persistent. Doing research is challenging, being optimistic, confident and persistent can help overcome various kinds of difficulties.
Being Resourceful: doing research is not just doing experiments. More importantly, it often requires you to deploy good tactics to jump out of the box and win strategically.
Being Flexible/Prepared to Discover, Create and Exploit Opportunities: an opportunity sometimes implies a shortcut to success.
Maintaining a Good Relationship with Other People: everyone has his/her own strengths and weaknesses. Build your success on the strengths of yourself and your partners.
References:
[1] Dongyuan Zhan’s Dissertation Research Description. http://cse.unl.edu/~dzhan/wiki/index.php5/Research
[2] Tilera's Balancing Act: 100 Cores vs. Market Realities. http://news.cnet.com/8301-13512_3-10388025-23.html
[3] Holland Computing Center. http://hcc.unl.edu/main/index.php