These days, heavy-duty computing is being performed less on local systems, and more with resources physically concentrated in remote data centers, which can handle these tasks much more efficiently. These arrangements are generally known as cloud computing, and they offer their customers many advantages, such as flexibility, scalability, and availability. Above all, they remove from users the need to worry about the physical ICT infrastructure. On the other hand, people who work in this area often face challenges bordering on the nightmarish: they must handle many conflicting requests in terms of bandwidth, memory allocation, number of central processing units (CPUs), and so on, all at once. This paper proposes a methodology to somewhat ease this burden.
The authors focus in particular on reducing latency sensitivity, or how the time between a database query and its response varies according to the overall system workload. The paper starts by describing the methodology in terms of system architecture, detailing in particular node deployment, or how to map logical applications to cloud instances; search techniques, or how to find nodes with the shortest distance among them in the shortest time; and network distance measurement, or how message size impacts communication speed. Then, it puts the methodology to work with different workloads and measures the results regarding latency. Finally, it reports about ongoing work in this area by other researchers.
Although this is an academic paper that will mostly be appreciated in that environment as a means to keep abreast of the latest developments in the field, it can also be appreciated by practitioners in data centers as a pointer to the most sensitive issues to watch in order to increase performance.