Sort Benchmark

Sort Benchmark.

New: We will soon release the specification for a new sort benchmark, CloudSort, that measures the total cost of ownership for external sorts performed in a cloud environment. In order to encourage entries for CloudSort, the deadline for the 2014 sort contest will be extended to October 1, 2014.

Background

Until 2007, the sort benchmarks were primary defined, sponsored and administered by Jim Gray. Following Jim’s disappearance at sea in January 2007, the sort benchmarks have been continued by a committee of past colleagues and sort benchmark winners. The Sort Benchmark committee members include:

Chris Nyberg of Ordinal Technology Corp
Mehul Shah of Amiato
Naga Govindaraju of Microsoft

Common Rules

All the sort benchmarks share the following ground rules:

Must sort to and from operating system files on secondary storage.
No raw disk usage allowed since we are trying to test the IO subsystem.
File or device striping (RAID 0) are allowed (encouraged) to get bandwidth. If file striping is used then the concatenated files must form a sorted file.
The output file must be created as part of the sort.
Time includes the launching of the sort program.
The sort input records must be 100 bytes in length, with the first 10 bytes being a random key.
Use the gensort record generator to create the input records.
The sort output file must be validated for correct key order and checksum.
The hardware used should be commercially available (off-the-shelf), and unmodified (e.g. no processor over or under clocking).

Link to the Sort Benchmark Home Page.