On Wed, Jul 15, 2009 at 9:18 AM, Alex Goncharov
<alex-goncharov@comcast.net> wrote:
,--- You/Suvankar (Wed, 15 Jul 2009 18:32:12 +0530) ----*
| Yes, I have got 2 segments and a master host. So, in a way processing
| should be faster in Greenplum.
No, it should not: it all depends on your data, SQL statements and
setup.
In my own experiments, with small amounts of stored data, PostgreSQL
beats Greenplum, which doesn't surprise me a bit.
Agreed. You're only operating on 99,000 rows. That isn't really enough rows to exercise the architecture of shared-nothing clusters. Now, I don't know greenplum very well, but I am familiar with another warehousing product with approximately the same architecture behind it. From all the testing I've done, you need to get into the 50 million plus row range before the architecture starts to be really effective. 99,000 rows probably fits completely into memory on the machine that you're testing PG with, so your test really isn't fair. On one PG box, you're just doing memory reads, and maybe some high-speed disk access, on the Greenplum setup, you've got network overhead on top of all that. Bottom line: You need to do a test with a number of rows that won't fit into memory, and won't be very quickly scanned from disk into memory. You need a LOT of data.
--Scott
