Performance

Please note that historical data is shown in this document. Results of more complete benchmarks will be published with later versions. Performance results for total execution times less than a few tens of seconds should be considered inaccurate.

Machines

Performance measurements for the following machines are available

For details on Amazon EC2 instances see Amazon EC2 Instance Types and Amazon EC2 Pricing

Master molecule store creation

Please note that the size discrepancy of the zinc-all dataset (16M vs 22M) comes from using an older version of the dataset with the i7-3610 measurement.

Fingerprint calculation

Most similar search performance

These measurements were made using the searchStorage tool, all measurements used 1024 bit binary fingerprints. All search used MOSTSIMILAR search mode. Search speed is the average query-target comparison speed. Search time does not include file read time.

Startup performance

Typical startup performance for reading molecular descriptors from large binary files was measured around 0.5 M/s - 1.5 M/s across various machines. This translates to several 10s of MB/s sustained read speed.

Memory requirements

Current implementation keep every relevant data in memory, thus the memory requirements are related to the number of structures, fingerprints and chosen fingerprint length. For large sets, using a single 1024 bit binary fingerprint as the descriptor the required memory of the current implementation can be estimated as 250-350 MB / million structures handled. This translates to the handling capacity of around 3-4 M structures per GB of available memory.