Hello PGSQL fans,
Looking back at my posts the past couple of days and the replies that I've got, 
I realized that I have failed to make one point clear: we are very pleased with 
what we have seen from PostgreSQL so far. Let me explain. At this point of 
developing or porting a benchmark on a new DBMS, the team usually deals with 
stability, scalability, or fundamental performance issues. Our fear was that 
working with an open source DBMS, we'd experience more issues than usual. But 
we got the kit running transactions on PGSQL quickly, and after some early 
tests, I decided to try the kit on a larger testbed (two other folks are the 
developers of the benchmark code; I design, run, and analyze the experiments). 
I have the benchmark running on a 300,000-customer database on a 16-CPU system, 
unusual for this early in the prototyping phase. People who developed TPC-E 
(the father of our benchmark) did their prototyping on commercial databases 
with much smaller databases on smaller systems. On this large testbed, PGSQL 
has been working like a champ, and performance is what I would call decent. Put 
in other words, I have been pleasantly surprised by the throughput I am getting 
out of the system, saturating a 16-way with no visible signs of contention when 
we reduce the database size.

We are developing a "reference" kit. People are not obligated to use it to 
publish official results. They can use it to kick the tires, then go to one of 
the commercial DBMS vendors and ask for their kit for an official TPC-V 
publication. Even if that's all that people do with the reference kit, our team 
has achieved the goal that the TPC set for us. What I am trying to do is see if 
we can take this to the point that people use PGSQL to publish official results 
and use it in competitive situations. It looks possible, so I'd love to see it 
happen.

Again, overall, our experience with PGSQL has been positive, even in terms of 
performance.

Thanks,
Reza

Reply via email to