Rapidly changing environments such as robots, sensor networks, or medical services are emerging. To deal with them, DBMS should persist sensor data streams instantaneously. To achieve the purpose, data persisting process must be accelerated. Though write ahead logging (WAL) acceleration is essential for the purpose, only a few researches are conducted.
To accelerate data persisting process, this paper proposes
remote WAL with asynchronous checkpointing
technique. Furthermore this paper designs and implements it. To evaluate the technique, this paper conducts experiments on an object relational DBMS called KRAFT.
The result of experiments shows that remote WAL overwhelms performance disk based WAL. As for throughput evaluation, best policy shows about 12 times better performance compared with disk based WAL. As for logging time, the policy shows lower than 1000 micro seconds which is the period of motor data acquisition on conventional robots.