qpid-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Andy Grove" <a...@codesuccess.com>
Subject RE: QPID M2 and Berkley DB
Date Tue, 13 Nov 2007 08:45:15 GMT
Carl,

Sorry for the slow response. I'm working part time on this particular
project at the moment. Yes, I would be very interested to see some numbers.
I'm more interested in scalability than raw performance per client. I'm
currently seeing maximum throughput of around 12,000 messages per second for
the particular test that I am running (sending 1k persistent transactional
messages to durable subscribers on a single topic with the publisher, broker
and consumer each running on separate quad-core 2.4ghz Intel boxes running
Fedora 8). Despite the good performance Java QPID/BDB is currently the
bottleneck in this application and I would like to be able to test the C++
broker. 

Thanks,

Andy.

-----Original Message-----
From: Carl Trieloff [mailto:cctrieloff@redhat.com] 
Sent: 05 November 2007 02:08
To: qpid-users@incubator.apache.org
Subject: Re: QPID M2 and Berkley DB


Andy,

In a few weeks it might be worth giving the C++ broker with the async 
store a run. We are still
tracking some bugs but are able to beat bdb performance by orders. If 
you like I can post some
numbers later this week.

Carl.

Robert Greig wrote:
> On 04/11/2007, Andy Grove <andy@codesuccess.com> wrote:
>
>   
>> Sure. That makes sense. In fact, I'm working on a product/solution where
>> I'll be using messaging and databases in a transactional way and it's
>> imperative that the performance of messaging (with persistent messages,
>> durable subscribers and transactions) is faster than the database
>> operations. I'm in the process of performance testing a prototype of the
>> solution and I'll be using QPID M2 and Berkeley DB for now and will see
how
>> the performance looks.
>>     
>
> It will be interesting if you could share your findings with us. In
> our performance testing (admittedly with a SAN) we have seen around
> 700 msgs/second (single commit per message).on a simple point to point
> test case.
>
>   
>> I don't have experience of HOWL but have worked with low-level
transaction
>> code in the past (implementing an EJB container). I guess there's no
reason
>> why I couldn't just have a go at creating an implementation of the
>> MessageStore interface - that would at least be a fast way to familiarise
>> myself with the work involved and see if it makes sense for me to
contribute
>> towards this.
>>     
>
> Yes definitely. The BDB store is probably a good example to follow.
>
> RG
>   



Mime
View raw message