The No Partitioning Optimized Join Implementation

Typedefs
typedef struct arg_t	arg_t

Functions
void	allocate_hashtable (hashtable_t **ppht, uint32_t nbuckets)
void	destroy_hashtable (hashtable_t *ht)
void	build_hashtable_st (hashtable_t *ht, relation_t *rel)
int64_t	probe_hashtable (hashtable_t *ht, relation_t *rel, void *output)
result_t *	NPO_st (relation_t *relR, relation_t *relS, int nthreads)
void	build_hashtable_mt (hashtable_t *ht, relation_t *rel, bucket_buffer_t **overflowbuf)
void *	npo_thread (void *param)
result_t *	NPO (relation_t *relR, relation_t *relS, int nthreads)

Detailed Description

Typedef Documentation

arg_t

typedef struct arg_t arg_t

Definition at line 88 of file no_partitioning_join.c.

Function Documentation

allocate_hashtable()

void allocate_hashtable	(	hashtable_t **	ppht,
		uint32_t	nbuckets
	)

Allocates a hashtable of NUM_BUCKETS and inits everything to 0.

Parameters

ht	pointer to a hashtable_t pointer

Not an elegant way of passing whether we will numa-localize, but this feature is experimental anyway.

Definition at line 183 of file no_partitioning_join.c.

build_hashtable_mt()

void build_hashtable_mt	(	hashtable_t *	ht,
		relation_t *	rel,
		bucket_buffer_t **	overflowbuf
	)

Multi-thread hashtable build method, ht is pre-allocated. Writes to buckets are synchronized via latches.

Parameters

ht	hastable to be built
rel	the build relation
overflowbuf	pre-allocated chunk of buckets for overflow use.

Definition at line 427 of file no_partitioning_join.c.

build_hashtable_st()

void build_hashtable_st	(	hashtable_t *	ht,
		relation_t *	rel
	)

Single-thread hashtable build method, ht is pre-allocated.

Parameters

ht	hastable to be built
rel	the build relation

Definition at line 231 of file no_partitioning_join.c.

destroy_hashtable()

void destroy_hashtable

(

hashtable_t *

ht

)

Releases memory allocated for the hashtable.

Parameters

ht	pointer to hashtable

Definition at line 218 of file no_partitioning_join.c.

NPO()

result_t * NPO	(	relation_t *	relR,
		relation_t *	relS,
		int	nthreads
	)

Definition at line 592 of file no_partitioning_join.c.

NPO_st()

result_t * NPO_st	(	relation_t *	relR,
		relation_t *	relS,
		int	nthreads
	)

The No Partitioning Join Optimized (NPO) as a single-threaded implementation. Just returns the number of result tuples.

Parameters

relR	input relation R - inner relation
relS	input relation S - outer relation

Returns

number of result tuples

Definition at line 357 of file no_partitioning_join.c.

npo_thread()

void * npo_thread

(

void *

param

)

Just a wrapper to call the build and probe for each thread.

Parameters

param

the parameters of the thread, i.e. tid, ht, reln, ...

Returns

Definition at line 492 of file no_partitioning_join.c.

probe_hashtable()

int64_t probe_hashtable	(	hashtable_t *	ht,
		relation_t *	rel,
		void *	output
	)

Probes the hashtable for the given outer relation, returns num results. This probing method is used for both single and multi-threaded version.

Parameters

ht	hashtable to be probed
rel	the probing outer relation
output	chained tuple buffer to write join results, i.e. rid pairs.

Returns

number of matching tuples

Definition at line 280 of file no_partitioning_join.c.