Re: Perpetuating the myth...annoying - Mailing list pgsql-advocacy

On 6/10/2004 7:10 AM, Kaarel wrote:
> scott.marlowe wrote:
>
>>I think that if we are going to benchmark the two, and I'm all for it, we
>>should benchmark them running something fairly standard, like the OSDL
>>tests.  And MySQL should be using innodb tables for everything, since
>>postgresql pretty much always carries the load of running mvcc and write
>>ahead logging and transactions for everything it does.
>>
>
> The other way, well not as good of course, would be to test PostgreSQL
> fsync turned off against MySQL MyISAM. That wouldn't still be complitely
> fair (unless there is a way to turn off implicit transactions of
> PostgreSQL) though but the results could be of interest in comparing
> PostgreSQL with default MySQL setup.

I have signed up for a project on PgFoundry (waiting for approval) to
make my PHP implementation of TPC-W available. It is ready ported to
MySQL 4.1.1 and runs of course on top of PostgreSQL >= 7.4.

It does take the differences in database functionality into account.
When running on PostgreSQL, it uses set returning stored procedures. The
same functionality is implemented in PHP of course when running on
MySQL. This is completely according to the TPC-W specifications since
the combination of Web Server(s) and Database is defined as System Under
Test and what counts is the response time to HTTP requests, not how fast
a simple select can spit out a row. The concurrency and transactional
requirements of that test suite make InnoDB tables required anyway, and
using pgpool PostgreSQL 7.4 currently scales better than MySQL 4.1.1.
But I am no MySQL expert, so there might be more tuning or some query
optimization possible ... right now the two are close enough together.

I think this is an interesting test, because it uses what a broad
variety of MySQL based applications are written in - PHP. And it
compares both databases on a fair level, by implementing the requested
business logic with what is available, instead of tayloring the features
or tests to the limited capabilities of the database.

The only thing missing to be 100% spec compliant woult be the remote
payment gateway emulator (credit card authorization simulation). That
would be some external server the PHP application during the order
processing has to communicate with via SSL.

Darcy Buskermolen did get it running and I have it currently in a
longterm test to check Slony Replication and will soon use it to develop
semi-automatic failover procedures for webservers using pgpool.


Jan

--
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck@Yahoo.com #