Re: lies damn lies and benchmarks

"Daniel Morgan" <dmorgan_at_exesolutions.com> wrote in message news:3CDABFF2.A8D28BE6_at_exesolutions.com...
>
>
> Pablo Sanchez wrote:
>
> > snipped for brevity>
> > > I am talking about the Microsoft employees in Redmond WA not the
> > Sybase
> > > team.
> >
>
> > Let me try a different tact: what pieces of SQL Server do you
believe
> > these employees develop? Let me know also if you believe that the
> > core team of Oracle kernel developers are still there who
developed,
> > let's say, read consistency.
>
> I can not reveal information I received after signing a
non-disclosure
> with respect to what they did at Microsoft.

Right.

>
> > I believe you're wrong on both these points. Clearly the TPC-C's
show
> > that SQL Server/Sybase don't have scalability issues. Both SQL
Server
> > and Sybase have row-level locking. What's your perceived issue
with
> > that?
>
> You are going to great lengths to avoid the dead elephant in the
middle of
> the room. Yes they have row level locking. My mother has row level
> locking.

Does she have a more cohesive argument base? :o

> They also still have lock escallation and page level locking as
> the number of row level locks is a limiting resource.

Prove that this is an issue with unbiased, factual data - no generalizations please. I won't respond until you do.

> > Multiversioning doesn't give you scalability. A well designed
> > application gives you scalability. Period.
>
> Mutliversioning isn't about scalability it is about consistency and
> concurrency.

Yes and no. Multiversioning and an in-memory locking scheme are both 'about consistency and concurrency' and as TPS increases, they have to scale. Both methods of implementing C&C have pro's and con's as already listed in a previous post.

Is one better than the other? Nope. TPC-C's and many Sybase ASE/SQL Server applications prove otherwise.

If you cannot provide unbiased data to the contrary, I submit this thread to be dead. Actually, since you haven't provided _any_ factual data on this thread to date, I'll kill the thread now....

To summarize, the two issues you pointed out: row level locking and multiversioning are non-issues.

--
Pablo Sanchez, High-Performance Database Engineering
mailto:pablo_at_hpdbe.com
http://www.hpdbe.com
Available for short-term and long-term contracts