A Survey of Join Processing in Data Streams

1. Introduction Given the fundamental role played by joins in querying relational databases, it is not surprising that stream join has also been the focus of much research on streams. Recall that relational (theta) join between two non-streaming relations R1 and R2, denoted RlweR2, returns thesetofallpairs (rl, r2), whererl E R1, 7-2 E R2, and the join condition 8(rl, r2) evaluates to true. A straightforward extension of join to streams gives the following semantics (in rough terms): At any time t, the set of output tuples generated thus far by the join between two streams S1 and S2 should be the same as the result of the relational (non-streaming) join between the sets of input tuples that have arrived thus far in S1 and sz. Stream join is a fbndamental operation for relating information from different streams. For example, given two stream of packets seen by network monitors placed at two routers, we can join the streams on packet ids to identify those packets that flowed through both routers, and compute the time it took for each such packet to reach the other router. As another example, an online auction system may generate two event streams: One signals opening of auctions and the other contains bids on the open auctions. A stream join is needed to relate bids with the corresponding open-auction events. As a third example, which involves a non-equality join, consider two data streams that arise in monitoring a cluster machine room, where one stream contains load information collected from different machines, and the other stream contains temperature readings from various sensors in the room. Using a stream join, we can look for possible correlations between loads on machines and temperatures at different locations