Motivation for System Research

(This is the closing speech of my lecture at the University of Science and Technology of China in November 2017, copied from Einstein’s “Motivation for Exploration”)

Most research in the field of systems can be divided into two categories: one is the emergence of new hardware, such as our programmable network cards, RDMA network cards, NVMe and NVM in the field of high-speed storage, SGX, TSX instruction extensions in CPUs, which can bring many new possibilities to system design; the other is the emergence of new application scenarios, such as deep learning that we are all talking about today, which brings many new challenges to system design. However, if there are only these two types of research in the field of systems, it will not become a respected research field, just as there can be no forest with only grass. Because for new hardware or new application scenarios, even if there are no scientists who specialize in system research, engineers will come up with ways to utilize these possibilities and meet these challenges.

So what attracts so many smart people into the field of system research? I think Einstein’s explains it well, so I’ll paraphrase it here. The first is a negative motivation, as Schopenhauer said, one of the strongest motivations that lead people to art and science is to escape the vulgar and despairing boredom of everyday life, to break free from the shackles of their own capricious desires. In engineering projects, there are always many non-technical factors, many historical legacy issues, many tool and hardware bugs. Most of the time in an engineering project is spent on these things that do not require much creativity, and the current AI is not enough to help us do well, so well-cultivated system engineers will want to escape this complexity and enter a world of objective perception.

In addition to this negative motivation, there is also a positive one. People always want to draw a simplified and easy-to-understand picture of the world in the most appropriate way; so he tries to replace the world of experience with his own world system and conquer it. This is what painters, poets, speculative philosophers, and natural scientists do, each in their own way. System researchers must strictly control the subject of their research, that is, to describe the most common modules in real systems. Attempting to reproduce complex systems in the real world with the precision and completeness of system researchers is beyond human intelligence. The basic abstractions that form the foundation of the system, such as IP for networks, SQL for databases, and files for operating systems, should be universally valid for a wide range of hardware architectures and application scenarios. With these basic abstractions, it is possible to construct a complete system through pure deduction. In this construction process, engineers can add the complexity of the real world, which may lose some of the good properties of the basic abstraction, but we can still understand the behavior of the entire system through a deductive process that does not exceed human reason.

The highest mission of system researchers is to obtain these universal basic abstractions, from which high-performance, scalable, highly available, and easy-to-program systems can be established by deductive methods. There is no logical path to these basic abstractions, only through intuition based on understanding of experience. This means that a good system researcher must first be an experienced system engineer. Because of this methodological uncertainty, it can be assumed that there will be many equally valid system abstractions. This view is valid both theoretically and practically. However, the development of the system field shows that at the same time, under the same hardware constraints and application scenarios, there is always one that seems much better than the others. This is what Leibniz very aptly described as “pre-established harmony”. The desire to see this pre-established harmony is the source of the infinite perseverance and patience of system researchers.