Re: [OJB-developers] "performance" target results against MySql

[Thomas M. <tho...@ho...>]

Hi Ricardo,

<snip>

Your performance results look really bad. To give you an impression that 
  this is not typical for OJB I have attached a zipped html document 
containing 3 tables.
The tables contain:
1. your testresults
2. my test against MS Access
3. my test against HSQLDB

The Result for HSQLDB are a bit distorted as HSQBL does most things 
inmemory. That's why it is so fast. But that's also why the OJB version 
is not that much slower (in absolute numbers) but the ratio is still 45%

For MS-Access you see a more "classic" OJB effect. Native JDBC is not 
very fast. OJB is slower. But at a very reasonable ratio: 19%

My experience with DB2 and ORACLE are similar, but I currently have no 
numbers available.

That's why I'm really shocked by your tests.


I have only a few possible explanations:
1. There have been changes between 0.7.314 (the version I used) and 
0.7.343 (your version) that had major impacts on performance. I had some 
few runs on my machine that show that this isn't the case.

2. The JDBC driver. People have often told me how fast MySql is. But 
when I see the results of your native JDBC tests I think it's rather slow...
Maybe that other internal problems of the MYSQL JDBC driver lead to such 
exorbitant results.
Maybe OJB uses certain features that are not well supported.

Maybe it's a good idea to share experiences with other MYSQL / OJB users 
to share their findings and insights. Everyone is invited to post his 
performance test results!

I am a bit surprised about the bad performance for primary key based 
look up (querying1 and 2). Are you sure there is a primary key defined 
for the table ?

3. OJB has become slow. As you can see from my tables OJB is not "slow" 
in general. OJB has been carefully performance tuned in the past.
But we have implemented a lot of new features that almost all have 
performance impacts.

After finishing all 1.0 features I want to have a complete code 
walkthrough to implement performance improvements.


> 
> 
> The results show that direct JDBC is 62.7% faster than OJB (against MySql):
> 
>       [ojb] Time: 1,393.023
>      [jdbc] Time: 874.658
> 
> 
> I think I found my problem....
> 
> :-(
> 



Mh, as mentioned above I don't believe that OJB is the only problem.

I'd like to discuss about aceptable performance. Which factor would be 
acceptable for you.
(For me even the native JDBC variant is definitely too slow).
20%, 40% ?

It's absolutely clear that an O/R layer (regardless which concrete 
product) *will allways* be slower as handcoded JDBC.

You get a lot of programming ease and pay it with performance. Period.
Of course we will work hard to minimize this overhead.

I have written the performance testsuite to make people aware of the 
performance impact they have for their specific environment.

I'm planning to provide a document on the performance TestSuite to help 
people understand all the numbers.

best regards,

Thomas