ROOT Bitmap Index Performance Result Kurt Stockinger KStockingerlbl

ROOT Bitmap Index Performance Result Kurt Stockinger <KStockinger@lbl. gov> 29 April 2005 ROOT-CORE Team meeting 22 April 2005 1

From a mail from Kurt Find attached some performance results of ROOT's TTree. Formula vs two different kinds of bitmap indices (equality encoding [BMI-EE] and range encoding [BMI-RE]. I used the same kind of queries as we did in December with Helmut. The data set consists of 10 million random floats with 10 attributes each. As you can see, the range encoded bitmap index (BMI-RE) is between a factor of 3 and 4 faster than TTree. Formula. The difference between the two bitmap indices is that equality encoding compresses better but is less efficient for query processing. Range encoding does not compress so well, but is query efficient for query processing. What is more, the query processing time is independent of the query selectivity. You can see that as the selectivity increases, equality encoding becomes less efficient whereas range encoding is still very efficient. We have also a three index strategy which is kind of a trade-off between these two. I will send you some more performance results next week. ROOT-CORE Team 22 April 2005 2

Kurt’s mail suite I've updated the code and included two Benchmark files (Benchmark 0. cpp, Benchmark. cpp) and the script that runs the queries. If you want to run the script, you need to change the paths for the data accordingly. Also note that the script unmounts the file system to guarantee a cold cache. The code can be downloaded from: http: //sdm. lbl. gov/~kurts/download/root-fastbit-v 0. 6. tar. gz Would you have some HEP data with some 10 millions records that we can use for further benchmarks? We have some data from Babar but it would be interesting to get some (simulated) LHC data as well. ROOT-CORE Team 22 April 2005 3

Tutorial output ROOT-CORE Team 22 April 2005 4

Just in: First Results with Ba. Bar Data ROOT-CORE Team 22 April 2005 5

Just in: First Results with Ba. Bar Data ROOT-CORE Team 22 April 2005 6