Hi All, We have implement a Siddhi Window using bloom filters to support large joins.
We did several test runs for getting the correlation between 'throughput' (million events per second) and 'false positive rate' for finding an optimal false positive rate for different window sizes when given a specific match rate (defines the number of matched events sent to the Siddhi Window). We have included the stats for each test run below and attached the result sheet for further reference [1]. According to the results, it seems that false positive rate varies with match rate for window sizes but not in an unique pattern. Since we need to input a false positive rate when creating a bloom filter, should we make it configurable when defining the Window? Please suggest a suitable configuration to choose an optimal false positive rate. So then we can continue more testing by changing some other parameters to find out the relations and best possible values. [1] https://docs.google.com/a/wso2.com/spreadsheets/d/1Gkgqk9qTbXswJyrrKTi2aCL2GpQw8JhceuxfarxP0ks Regards, -- *Dhanushka Priyasad* Software Engineering Intern WSO2, Inc. lean . enterprise . middleware Mobile: +94 713317472
_______________________________________________ Dev mailing list [email protected] http://wso2.org/cgi-bin/mailman/listinfo/dev
