Sure, that makes sense.
Just in current enviroment it may be runaway query,
which will take days to execute for some engines…

Thanks for you answer. I see your point, though the problem is that queries that do not have any restrictions are essential for answering the following typical business questions (do not know whether they can be answered with the data in your example database):
- How many visits have we had so far in total?
- How many pages has been created?
- What is the distribution of visits per page?
- How has the development of page creations been?

All these sorts of questions run on all the data. I believe that looking at fractions of data is more the IT-perspective on data.

I added your query, but I’d like to have at least date restriction,
as we may have different amount of data loaded into fact table, depending on scale factor.

This query is important to me because it look at ALL the data in the fact table without any joins. This sort of queries gives an overview of all data, which is typical in my field of data mining. I suspect that row-based engines will perform better for this one.

I *wish* I was (and hope to be someday) that expert.

Sounds great! What kind of box will you be using and how much memory/nr of disks will it have? E.g. Infobright compresses data quite aggressively so your 172GB will probably fit in 64 or even 32GB of memory, and Greenplum likes lots of disks (at least 1 data segment per CPU core, so with a dual quadcore machine you get better results with 8 partitions on separate drives). What you should also try to do is simulate concurrent access by firing multiple concurrent query streams. Some database are fast when running one query at a time but get crippled when the have to perform multiple tasks simultaneously.

About the ‘holy wars’ you mention: look at http://bit.ly/caLeRF where I ran TPCH using Greenplum and got discussions between PostgreSQL and GP advocates…

anyway, good luck with the benchmark and if you need help (or access to my machine , just let me know.

best, Jos

There are two factors:
1) I am not expert in PostgreSQL
2) Results, which directly compare PostregSQL vs MySQL, usually cause holy wars, and one who presents them, is claimed unqualified a) in PostgreSQL, b) in MySQL, c) in both.
So I do not want to go that way.

However if there is expert who wants to install PostgreSQL, tune it, and run proposed scenarios,
I would be happy to even provide access to box.

I suggested what I wrote before because someone actually might want to do some analytics on a cms or an e-shop – visits, visitors, countries, extensive products history, etc.
I see your point of it being perhaps too lengthy to bother with it, though.

I do not think db design in CMS suitable for OLAP benchmarks.
Either I do not think OLTP oriented FireBird and PostgresSQL will be good for proposed benchmark.

I expect that queries will take 5-1000 mins to execute, and if we do not use special engine,
it will be just pain. I am going to ran on MySQL MyISAM / InnoDB only to have baseline.