Abstract:

The traditional data stream processing systems will keep all the tuples in the memory until they have been processed in order to provide reliable tuple processing.Unfortunately,the strategy will take up much memory.To address this issue,a cost-effective approach to tracking tuples-TTDSP was proposed.The approach includes three strategies,namely memory allocation strategy,tuple acker selection strategy and checksum updating strategy,which make tuple acker to keep only the XOR checksum not the tuple in memory.Moreover,the tuple acker are load balancing through the improved consistent Hash.The experiments on memory overhead and load balancing show that this approach is able to track and process tuples effectively and reliably.

Key words: data stream processing, reliability,data stream, load balancing