[Mulgara-general] Re: [Mulgara-dev] The future of Mulgara. What is 'XA2' and how will it benefit you.

Andrae Muys andrae at netymon.com
Wed Nov 29 12:07:18 UTC 2006


On 29/11/2006, at 5:44 PM, Tate Jones wrote:

> An accurate understanding of the problem.  From memory Kowari was  
> heavily IO bound after 250 million.
>
> Some additional thoughts for the document
>
> - Isolation levels - should XA2 support phantom and repeatable  
> reads? Configurable at the session transaction level?

The proposed isolation level is higher than 'no_phantom_reads', but  
falls just short of 'serialisable'.  I am almost convinced that  
serialisable cannot be obtained without locking.

OTOH, applications requiring RDFS/OWL consistency semantics will  
require serialisability, so there I am proposing that we should  
eventually provide a lock-service, that will provide the level of  
guarantees RDFS/OWL applications require (provided no application by  
passes the lock by using RDF semantics directly).

I still have some work to do finishing the documentation for the  
transaction rearchitecture.  When that's done my top priority is  
preparing a paper/seminar to cover these issues in greater depth.

> - Simple test plan (how to confirm the scalability, performance and  
> robustness)

Robustness is hard to 'test'.  I have previously proposed a move to  
crash-only shutdown as a means to provide some of this.  Ultimately  
robustness will be a matter of specifying robust semantics, and  
providing some sort of verification (probably manual inspection).

Scalability and performance are easier to measure.  Insert speed is  
naturally important for various varieties of dataset.  Some benchmark  
queries will also be required.  What figures of merit are people  
interested in?

> On 27/11/2006, at 11:27 AM, Andrae Muys wrote:
>>
>> http://www.netymon.com/papers/XA2-Discussion-Paper.pdf

-- 
Andrae Muys
andrae at netymon.com
Principal Mulgara Consultant
Netymon Pty Ltd





More information about the Mulgara-general mailing list