Comments on: Extreme Buzz About Oracle OpenWorld

By: ghassan

ghassan — Sun, 21 Sep 2008 17:50:35 +0000

Brett,
TimesTen is not intended for DWH, so it is not easy to compare it to ParAcell or other such databases.
As for what Oracle will announce, well wait and see.

By: Brett Schroeder

Brett Schroeder — Sat, 20 Sep 2008 22:32:33 +0000

Thanks for the paper reference.

Since Exasol, ParAcell etc are “in-memory” databases, then wouldn’t a more realistic/fair comparison be against Oracle Times Ten DB (instead of 10g or 11g)?

By: Greg Rahn

Greg Rahn — Fri, 19 Sep 2008 18:24:55 +0000

@Brett Unfortunately I can not comment on speculation but I would like to add some thoughts with regards to the rest of your comments. Vertica has not and can not run a TPC-H benchmark due to limitations of the current product. In order to use Vertica, you must have a star schema and the TPC-H schema is not a star schema, it most closely resembles a 3NF model. The reason for this is the Vertica query optimizer is very limited. Some limitations of the current optimizer are that it does not support subqueries or IN lists. It also does not support any of the analytic functions commonly found in BI/DW queries that Oracle does. Vendors like ParAccel and Exasol have achieved top spots in the TPC-H 3TB scale factor and under by approaching the problem differently. Their secret sauce is to have all the data in memory. I don't think it takes a computer scientist to tell you that reading data from memory will be an order of magnitude faster than reading data from disk. But what happens when they have to read data from disk? I suspect the results are not nearly as stunning. I would encourage you and others who think column store databases are superior to row stores for BI/DW workloads to look at Allison Holloway and David DeWitt's presentation from VLDB 2008 which has an alternate title of "Yet Another Row Store vs Column Store Paper".

By: Chen Shapira

Chen Shapira — Fri, 19 Sep 2008 17:49:28 +0000

I took the bait. Originally I planned to attend Real World Performance Roundtable, but it was dropped in favor of Database Accelerator.

Poor Juan Loaiza – he’ll be giving same presentation three times on Thursday!

By: Brett Schroeder

Brett Schroeder — Fri, 19 Sep 2008 14:41:22 +0000

My guess is that Oracle will get column-store capability (great for DSS/DW, not so great for OLTP). Products like Vertica claim to beat Oracle on (unaudited) TPCH by factors of 10-100 in query time. All the major RDBMS’s vendors are lagging new products like Vertica, Exasol, Netezza, Kickfire etc in warehouse performance. Consider these results for TPCH @ 1TB: Oracle on Solaris = 119,000 QphH and costs $24/QphH while an Exasol product = 1,000,000 QphH and costs $1/QphH….that’s 10X faster and 24X cheaper. Granted, Oracle does a lot of things these products don’t do but numbers like that force any company to think long and hard about buying more Oracle licenses.

Column store is something that Oracle can code up relatively quickly. In addition, Kevin Closson wrote several times about these new data warehouse products and column-store, then went awfully quiet on his blog :-) I suspect he’s had his head down coding and/or testing this column-store feature.