著者

Negishi Yasushi Kawachiya Kiyokuni Murata Hiroki TAGO Kazuya

出版者

一般社団法人情報処理学会

雑誌

情報処理学会論文誌 (ISSN:03875806)

巻号頁・発行日

vol.41, no.10, pp.2881-2894, 2000-10-15

---

Progress in semiconductor technology has made it possible to build small network clients compact enough to be embedded in credit cards or wallets. These devices, which have severely restricted computing resources, are called "micro-clients." We propose an approach for building systems for micro-clients. The requirements of a system for micro-clients are as follows : (1)It must work with a small amount of memory and a low-power processor. (2)It must work even while the link is disconnected, because of the high cost of communication. (3)It must work with low-quality communication links. We introduce a software system called Tuplink that meets these requirements. The Tuplink system on the client node manages a central data pool to hold user and system information in an integrated manner. The server node also has a data pool for each client, and the contents of both pools are kept identical by means of a communication network. This server-side data pool makes it easier to build a system in which necessary functions are divided between server and client nodes. One-to-one communication between client and server is abstracted by a synchronization operation between the two pools in order to hide link management and communication timing from other subsystems. A communication protocol for synchronization, called the Tuplink protocol, is used to synchronize the two pools. Other subsystems, such as the RPC, file, and detabase systems, use the central data pool instead of their own buffers. The state of the communication link is hidden from other subsystems. We call this the "meta-middleware" approach. The Tuplink system meets the above requirements as follows : (1)Use of the central data pool eliminates duplication of buffers among subsystems and copying of data among buffers. (2)The system can operate while the link is disconnected by using the central data pool as a daga cache. (3)The synchronization protocol efficiently handles packet loss by using the central data pool as a communication buffer. We have built systems based on the Tuplink model, and applications for them on several platforms, including a smart phone, Palm III, and Windows CE. This paper discusses the approach and findings of the system implementation.