50 likes | 120 Views
SWiM Panel on Engine Implementation. Jennifer Widom. Synopses. Running Op. Ready Op. Waiting Op. Query Processing Architecture. Output Stream. Query Plans. Applications register continuous queries. p. X. Users issue continuous and ad-hoc queries. s. s. X.
E N D
SWiM Panel on Engine Implementation Jennifer Widom
Synopses Running Op Ready Op Waiting Op Query Processing Architecture Output Stream Query Plans Applications register continuous queries p X Users issue continuous and ad-hoc queries s s X Scheduler Input Data Streams Administrator can monitor query execution and adjust run-time parameters
Issues • Memory versus disk for intermediate state • More generally: Is therea clear latency-accuracy-storage tradeoff? • What is the objective of the scheduler? • More generally: What are the performance metrics for continuous queries?
Issues (cont’d) • How important is a built-in approach to approximation? • More generally: Will data streams really be all that fast? If so, can’t we expect the source to filter? • How adaptive does execution really need to be? • More generally: Will data streams really be all that variable?
Issues (cont’d) • What does the API for a Data Stream Management System look like? • For registering queries • For seeing results • For understanding approximation • For monitoring the system