Difference between revisions of "Gsoc08-hashindex"

From PostgreSQL wiki
Jump to: navigation, search
m (Performance)
m (Performance)
Line 17: Line 17:
 
Memory = 1GB
 
Memory = 1GB
 
* Workload
 
* Workload
2000 single point query
+
2000 equality query
 
* Result
 
* Result
 
{| border="1"
 
{| border="1"

Revision as of 12:05, 17 August 2008

Proposal Of GSoC

Ideas

There's two basic ideas in the patch.

  • Store the hash value instead of real key in the bucket.

We can keep more tuples in a bucket and reduce the index size. It also means all hash indexscans become lossy and have to be rechecked at the heap.

  • Keep the contents of each index page ordered by hash value and use binary instead of linear search to find the matching item(s) during an indexscan.

Performance

You can see Details About The Test for more information.

  • Platform

Linux = 2.6.24-16-generic

Processor = Intel(R) Core(TM)2Duo CPU T7500@2.20GHz

Memory = 1GB

  • Workload

2000 equality query

  • Result
index build time index size query time blocks read
hash 504650.100 ms 1024MB 5189.844 ms 39801
btree 844495.867 ms 1027MB 5656.045ms 41627