This is a listing of successful results of all the various data storage and processing system benchmarks I've conducted using the dataset produced in the Billion Taxi Rides in Redshift blog post. The dataset itself has 1.1 billion records, 51 columns and takes up about 500 GB of disk space uncompressed.
Though these benchmarks used the same dataset, I wouldn't consider any two as comparing apples to apples. The times recorded don't necessarily reflect the top performance these systems are capable of. It's likely each system could have performed faster if more effort was put into setup, configuration and/or better hardware was available. There are some cases where I re-ran a benchmark with more tuning to achieve a better result and in these cases only the fastest benchmark will be listed below.
The table is sorted by the fastest time query 1 finished in (measured in seconds).
|Query 1||Query 2||Query 3||Query 4||Setup|
|0.021||0.053||0.165||0.51||MapD & 8 Nvidia Pascal Titan Xs|
|0.027||0.083||0.163||0.891||MapD & 8 Nvidia Tesla K80s|
|0.028||0.2||0.237||0.578||MapD & 4-node g2.8xlarge cluster|
|0.036||0.131||0.439||0.964||MapD & 4 Nvidia Titan Xs|
|0.051||0.146||0.047||0.794||kdb+/q & 4 Intel Xeon Phi 7210 CPUs|
|1.034||3.058||5.354||12.748||ClickHouse, Intel Core i5 4670K|
|1.56||1.25||2.25||2.97||Redshift, 6-node ds2.8xlarge cluster|
|4||4||10||21||Presto, 50-node n1-standard-4 cluster|
|8.1||18.18||n/a||n/a||Elasticsearch (heavily tuned)|
|10.19||8.134||19.624||85.942||Spark 2.1, 11 x m3.xlarge cluster w/ HDFS|
|11||10||21||31||Presto, 10-node n1-standard-4 cluster|
|14.389||32.148||33.448||67.312||Vertica, Intel Core i5 4670K|
|34.48||63.3||n/a||b/a||Elasticsearch (lightly tuned)|
|35||39||64||81||Presto, 5-node m3.xlarge cluster w/ HDFS|
|43||45||27||44||Presto, 50-node m3.xlarge cluster w/ S3|
|152||175||235||368||PostgreSQL 9.5 & cstore_fdw|
|264||313||620||961||Spark 1.6, 5-node m3.xlarge cluster w/ S3|