Thinking Big About Tiny Databases (UC Berkeley)

This paper introduces the current work on tiny database from UC Berkeley. 

Early “tiny databases” provide high-level languages to maintain data without worrying about low-level details regarding how the data is collected or processed. 

The latest (2007) refinements from TinyDB group include:

• Declarative network programming.

The UC Berkeley has implemented a P2 system, which is a overlay network uses a high-level declarative language to express overlay networks in a highly compact and reusable form.

Because of the difficulty of programming in sensor networks, the declarative sensor network (DSN) is proposed to simpler the program. It simply a) power management, b) network design, and c) data management. In the other hand, it is reusable. We can write the same code to each senor node and add it to an existing network quickly.

The open research issues including a) query optimization – they are designing a declarative meta-optimizer for P2 and DSN, b) heterogeneous sensornets, c) power management, and d) distributed statistical methods, e.g. Bayesian Belief Propagation. 

• Sophisticated signal-oriented programming languages.

Take the following acoustic application for an example, using FFT operator, we convert high data rate audio singles into the frequency domain to identify frequencies that are characteristic of the objects being tracked. When such frequencies are detected, we can apply beamforming algorithms to perform triangulation for tracking purpose.

The UC Berkeley has developed a programming language called WaveScript and a sensor networkruntime system called WaveScope. Under WaveScript and WaveScope, the programmer can perform complex signal and data processing operators on sensor data directly without considering details of networking protocols. 

• Support for large-scale sensing applications.

There is a so-called High Fan-in (HiFi) system which is programmable with a uniform, declarative, data-centric language. In this system, they develop a virtual device interface to avoid the complexity of the underlying devices. Also, HiFi will deal with the multiple queries in the hierarchical HiFi environment. 

• Support for mobility and intermittent connectivity.

Sensor networks are dynamical, i.e. there always are small changes in connectivity because of the nodes failure, work mode switch on/off. If it is a bad connectivity, the bandwidth of network will become smaller, the data transfer will be clogged. 

They build a new data management component called ICEDB (for Intermittently Connected Embedded Data Base) to solve the mobility issue. Through ICEDB, users can assign both local and global priority to determine the data transfer order. Under the control of different priorities, each data package can be transferred in a suitable serial.