Run your own TPC-H Benchmark with #Oracle or #Exasol

The TPC-H Benchmark is for Decision Support Systems. It’s described very detailed on the TPC.ORG site, but you may find it quite an effort to generate the data and prepare the SQL for table creation and reporting.

At least I did, which is why I thought having that all ready for download and run would be helpful.

What I have prepared for Oracle and for Exasol is:

  • The data files (CSV format) for the 1 GB TPC-H
  • The DDL for the TPC-H tables
  • The loader commands to populate these tables
  • The 22 queries for the TPC-H benchmark

You can download it here:

TPC-H for Oracle          TPC-H for Exasol

The data volume is of course quite small for a production data warehouse but ideal for quick testing and self-education. I’m using it together with VirtualBox and VMs on my notebook with 16 GB memory.

See here for a demo – I’m setting up the TPC-H for both Oracle and Exasol and then I do a comparison:

Some remarks about the comparison:

I’m an Exasol employee and the outcome is very positive for Exasol.

Never the less, I tried to do a fair comparison. It’s just running the pure 22 SELECT statements, no tuning, no tweaking of the Exasol database or the underlying VM.

The Oracle version is quite recent (18.3) but not the most recent, same with the Exasol version (6.2), not the just released Exasol 7.0.

As you can see, the Exasol database is out of the box about 6 times faster than the Oracle database for the same workload having the same hardware resources – without any tuning.

I suppose you could get better performance from Oracle for the 22 queries with some effort, like analyzing the workload, adding indexes of the certain available types, partitioning the tables, adding SQL Profiles and Optimizer Directives etc.

The point is, that’s all not required with Exasol. I just run the workload twice and everything is self-optimized afterwards.

You could call this an autonomous database 😉

It’s totally easy to reproduce the test for yourself: Just download our free Community Edition; it’s what I’m using in this benchmark.

Keep in mind that this is a Decision Support System benchmark with an analytical workload. Oracle looks much better in a comparison with an OLTP workload.

But for analytics: Exasol stands behind nobody.

To all the other vendor’s presales consultants out there who encounter us on a PoC: Good luck 🙂

,

  1. #1 by Clemens Bleile on August 11, 2020 - 13:00

    Hi Uwe,

    this test without indexes is just a question of disk speed (HDD or SSD) and caching (outside the Oracle cache). On my first test running on Virtualbox (4GB RAM provided for the VM) and Oracle 19.6. with

    FILESYSTEMIO_OPTIONS=’SETALL’
    SGA_TARGET=1G
    PGA_AGGREGATE_TARGET=2G

    it took 21.1 secs. Here the single results:

    Elapsed: 00:00:02.12
    Elapsed: 00:00:00.17
    Elapsed: 00:00:00.72
    Elapsed: 00:00:00.66
    Elapsed: 00:00:01.00
    Elapsed: 00:00:00.43
    Elapsed: 00:00:00.74
    Elapsed: 00:00:03.60
    Elapsed: 00:00:01.29
    Elapsed: 00:00:00.75
    Elapsed: 00:00:00.27
    Elapsed: 00:00:00.78
    Elapsed: 00:00:00.68
    Elapsed: 00:00:00.48
    Elapsed: 00:00:00.50
    Elapsed: 00:00:00.17
    Elapsed: 00:00:00.94
    Elapsed: 00:00:01.99
    Elapsed: 00:00:01.00
    Elapsed: 00:00:00.72
    Elapsed: 00:00:01.93
    Elapsed: 00:00:00.16

    Second test took 17.43 secs.

    Running with filesystem-caching (filesystemio_options=’ASYNCH’) the first test already took only 15.72secs.

    Regards
    Clemens

  2. #2 by Uwe Hesse on August 11, 2020 - 19:30

    Hi Clemens, thank you for publishing your numbers here! I have no doubts that Oracle can run the 22 queries in less than 60 seconds. The point I try to make is: The two databases run the same workload with the same hardware resources and Exasol is 6 times faster without any need to tweak the hardware or the database. You know, the two VMs in my test run physically on the same notebook at the same time. Your machine is apparently better than mine. What would make your test more meaningful would be if you would also run the Exasol benchmark with the same conditions. Maybe it’s done in 3 seconds?

  1. Comparison between #Oracle and #Exasol | Uwe Hesse

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: