Re: [osdldbt-general] Error when test on DBT2 Postgresql

[Mark W. <ma...@2n...>]

Hi Tim,

Sorry for the late reply.

On 11/10/15 05:08, Tim Chou wrote:
> Hi Mark,
> 
> Thank you for the reply.
> 
> It's very helpful.
> 
> 1. The attachment is my result when I run the test on 10 warehouses with
> 1000 connections and no thinking time.
> 
> 2. Where do you get the rule of thumb? I have read TPC-C documents.
> However, I didn't find anything about this.
> In my opinion, each transaction should be finished in 11s. If we assume
> there is no data contention in our environment, then the metric txns/min
> should be near 10 x # of warehouse x 60s / 11s, which is equal to 60 x # of
> warehouse. Is this right?

In v5.11 of the spec, it's in the comment for clause 4.1.3:

The intent of this clause is to prevent reporting a throughput that
exceeds this maximum, which is computed to be 12.86 tpmC per warehouse.

> 3. You can ignore my words in the third question. Sorry.
> 
> 4. Actually, I have another question about the benchmark. TPC-C has defined
> one warehouses has 10 terminals.
> So what's the meaning of connections?
> I think we can just create threads (10 x #warehouse ) to keep submitting
> the transactions until the test ends.

Ah, this is specific to the dbt2 kit.  When specifying the number of
warehouses, the number of clients/terminals emulated is always 10 *
warehouses.  The "connections" terminology refers to the number of
database connections that are opened.  The "dbt2-client" program is a
simple connection concentrator between the clients/terminals to the
database.

> 5. Now, I'm trying to conclude the difference between DBT2 and TPC-C
> official docs. If I have got something, I will post in our mailing list.
> 
> Thanks for your efforts on the benchmark. It's really helpful and useful.

I'm glad it is helpful for you. :)

Regards,
Mark

-- 
Mark Wong                                http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, RemoteDBA, Training & Services