Re: 8 hours tests ends with inconsistent DB.

[Jens Vagelpohl <jens@dataflake.org>]

> I apologize for not using the term "benchmark" correctly. One may have 
> different expectations about the result of a "benchmark":
>
> a) results
> b) comparable results
>
> if b) applies you are right about missing specs, but I guess the OP 
> just wanted to test the limits of OL and the issue was not poor 
> performance but corrupt DB and lack of stability under extrem load. 
> You cannot blame Sleepycat for that (bugs aside).

At Zope Corporation (-> Open-Source web application server Zope) we had 
several very intelligent people trying to develop a data storage 
backend for Zope using BerkeleyDB for a couple years. Just a few weeks 
ago that attempt was finally given up on. We were unable to work around 
all the BDB weirdnesses and a lot of the time it seems Sleepycat wasn't 
all that responsive when it came to bugfixes (several were uncovered by 
our work) or suggestions. I would not be surprised if at least some of 
those corruptions can be blamed squarely on BDB bugs.

jens