[Bizgres-general] Cutover to 8.1.3 merge in progress
Mark Kirkwood
markir at paradise.net.nz
Mon Mar 6 22:01:22 GMT 2006
I redid the TCP-H tests comparing Bizgres current (8.1.3 codebase) to
vanilla 8.1.3 (see below):
The details for the queries etc are in previous postings.
The results are looking pretty amazing(!) for index creation time and
size, retrieval time however looks pretty similar.
I suspect the retrieval time is being mainly spent in heap fetches, so
index access (bitmap or btree) is not the bottleneck (I can hear someone
muttering about how good index-only access would be....).
Anyway the results:
---------------------------------------------------------------------
Index creation time:
Bizgres current Postgres 8.1.3
(s) (s)
customer_c_custkey 9 9
customer_c_nationkey 26 13
customer_c_mktsegment 24 20
orders_o_custkey 28 272
orders_o_orderstatus 21 263
orders_o_orderpriority 24 292
Index sizes:
Bizgres current Postgres 8.1.3
(8K pages) (8K pages)
customer_c_custkey 3288 3288
customer_c_mktsegment 158 5779
customer_c_nationkey 334 3288
orders_o_custkey 32847 32847
orders_o_orderpriority 1428 65876
orders_o_orderstatus 717 41131
Queries
Bizgres current Postgres 8.1.3
(s) (s)
Customers
1 5 6
2 17 17
3 15 15
4 1 1
Orders
1 33 35
2 77 78
3 88 91
4 26 48
Customer Orders
1 45 42
2 44 51
3 48 53
4 33 39
-----------------------------------------------------------------------
Those of you that were following this will see that the results for
vanilla 8.1.3 are much better than vanilla 8.1.1 - I had a motherboard
failure and changed a Tyan S2510 for a Supermicro P3TDER and upgraded a
Promise TX2000 for a 3Ware 7506. The new setup is much faster so
unfortunately I've spoiled any comparison to the previous results.
regards
Mark
More information about the Bizgres-general
mailing list