GPU enabled database systems

1. A method for parallel query execution of a database operation on a database utilizing a graphics procession unit (GPU), the method comprising:

receiving a query by a host, the query including at least two database relations;

starting at least one GPU kernel of a plurality of GPU kernels, wherein the at least one GPU kernels include a GPU memory allocated at runtime;

hash partitioning the at least two database relations by the GPU kernel resulting in at least two hash partitioned database relations, wherein each of the at least two hash partitioned database relations includes a partition key;

loading the at least two hash partitioned database relations into the GPU memory;

loading the at least two hash partitioned database relations and partition keys corresponding with the at least two hash partitioned database relations into the GPU memory;

building a hash table for a smaller of the at least two hash partitioned database relations; and

executing the query by the at least one GPU kernel thereby providing in hash table GPU memory access, wherein the parallel query execution of the database is performed solely by the GPU.

2. The method of claim 1 further comprising returning a result of the query.

3. The method of claim 1 , further comprising:

when the query is a long query including a plurality of operators, parsing the long query into a plurality of sub-queries;

for each of the sub-queries, starting one of the at least one GPU kernels such that the sub-queries are processed in parallel; and

returning to the hash partitioning.

4. The method of claim 1 , wherein

the host is a GPU, and wherein

a database data and a plurality of database keys are stored on the GPU memory.

5. The method of claim 1 , wherein

the host is a GPU, wherein

a database data and a first plurality of database keys are stored on a disk data store, and wherein

a second plurality of database keys are stored on the GPU memory.

6. The method of claim 5 , further comprising:

synchronizing the first plurality of database keys with the second plurality of database keys.

7. The method of claim 1 , wherein

the host is a database node, wherein

a first database data and a first plurality of database keys are stored on a disk data store, and wherein

a second database data and a second plurality of database keys are stored on the GPU memory.

8. The method of claim 7 wherein the data disk store is pipelined with the GPU.

9. The method of claim 1 , wherein the database is configured for logging transactions such that a roll back may be initiated if an inconsistent state occurs between the host and the GPU.

10. The method of claim 1 , wherein the database is configured for limiting memory usage such that the query is restricted to a limited class of queries.

11. The method of claim 1 , wherein the GPU is a G 80 architecture compliant GPU.

12. The method of claim 1 , wherein 1 the GPU memory is a shared memory for the at least one GPU kernels.

13. The method of claim 1 , wherein the two hash partitioned relations are into the GPU memory in a coalesced access pattern such that a contiguous region of memory is utilized.

14. A computing device program product for parallel query execution of a database operation on a database utilizing a graphics processing unit (GPU) in coordination with a computing device, the computing device program product comprising:

a computer readable medium;

first programmatic instructions for receiving a query by a host, the query including at least two database relations;

second programmatic instructions for starting at least one GPU kernel of GPU kernels, wherein the at least one GPU kernels include a GPU memory allocated at runtime;

third programmatic instructions for hash partitioning the at least two database relations by the GPU kernel resulting in at least two hash partitioned database relations, wherein each of the at least two hash partitioned database relations, includes a partition key;

fourth programmatic instructions for loading the at least two hash partitioned database relations into the GPU memory;

fifth programmatic instructions for loading the at least two hash partitioned database relations and partition keys corresponding with the at least two hash partitioned database relations into the GPU memory;

sixth programmatic instructions for building a hash table for a smaller of the at least two hash partitioned database relations;

seventh programmatic instructions for executing the query by the at least one GPU kernel thereby providing in hash table CPU memory access, wherein the parallel query execution of the database is performed solely by the GPU; and

eighth programmatic instructions for returning a result of the query, wherein the programmatic instruction are stored on the computer readable storage medium.

15. The computer program of claim 14 , further comprising:

when the query is a long query including a plurality of operators, ninth programmatic instructions for parsing the long query into a plurality of sub-queries;

for each of the sub-queries, tenth programmatic instructions for starting one of the at least one GPU kernels such that the sub-queries are processed in parallel; and

eleventh programmatic instructions for returning to the hash partitioning.

16. The computer program of claim 14 , wherein

the host is a GPU, and wherein

a database data and a plurality of database keys are stored on the GPU memory.

17. The computer program of claim 14 , further comprising:

twelfth programmatic instructions for synchronizing a first plurality of database keys with a second plurality of database keys, wherein

the host is a CPU, wherein

a database data and the first plurality of database keys are stored on a disk data store, and wherein

the second plurality of database keys are stored on the GPU memory.

18. The computer program of claim 14 , wherein

the host is a database node, wherein

a first database data and a first plurality of database keys are stored on a disk data store, and wherein

a second database data and a second plurality of database keys are stored on the GPU memory.

19. A graphics processing unit (GPU) enabled database system for parallel query execution comprising:

a database;

a GPU in communication with the database, wherein

the GPU is configured for processing a database query, the processing including,

hash partitioning the at least two database relations by a GPU kernel resulting in at least two hash partitioned database relations, wherein each of the at least two hash partitioned database relations includes a partition key;

loading the at least two hash partitioned database relations into the GPU memory allocated at runtime;

loading the at least two hash partitioned database relations and partition keys corresponding with the at least two hash partitioned database relations into the GPU memory;

building a hash table for a smaller of the at least two hash partitioned database relations; and

executing the query by the at least one GPU kernel thereby providing in hash table GPU memory access, wherein the parallel query execution of the database is performed solely by the GPU, wherein

the GPU includes at least one GPU kernel of a plurality of GPU kernels, the at least GPU kernel including GPU memory allocated at runtime, and wherein

the GPU is configured for parallel processing the database query; and

a host for receiving database queries.

20. The system of claim 19 , wherein

the host is a CPU, and wherein

a database data and a plurality of database keys are stored on the GPU memory.

21. The system of claim 19 , wherein

the host is a CPU, wherein

a database data and a first plurality of database keys are stored on a disk data store, and wherein

a second plurality of database keys are stored on the GPU memory.

22. The system of claim 19 , wherein

the host is a database node, wherein

a first database data and first plurality of database keys are stored on a disk data store, and wherein

a second database data and a second plurality of database keys are stored on the GPU memory.