Idle-time processing in time-slide window join

Abstract

When memory is exhausted from bursty tuple arrivals, some portion of tuples in memory can be evicted to disk, and those tuples can be recalled to complement earlier results when a system becomes idle. This paper presents a method for the idle-time processing when sliding window join queries are given to deal with continuous data streams. Regarding this, we discuss (i) how to determine the priority of tuples for victim selection when memory is not enough and (ii) how to ensure that no result tuple is generated twice when join is conducted with disk-resident tuples in idle time.