Re: Probable faq: need some benchmarks of pgsql vr.s mysql

On Fri, Oct 29, 2010 at 1:25 PM, Josh Berkus <josh(at)agliodbs(dot)com> wrote:

> Brian,
>
>
> My company is currently hitting a problem with mysql/innodb having
>> really slow insert performance (we're seeing ~1K rows/second).
>>
>
> As others have said, there's clearly something wrong with your setup and/or
> application which would likely affect PostgreSQL as well. I can name several
> ways in which PostgreSQL is better than MySQL/InnoDB, but insert performance
> isn't one of them.
>
>
> My boss
>> wants to go and spend a bunch of money on the Tokutek backend.
>>
>
> I'm sure that Tokutek would be thrilled to have a customer (they're very
> new). Again, though, insert performance isn't Tokutek's specialty; the
> arguments for using them is that (a) they're not owned by oracle,

<snip>

That seems like a really weak argument, unless Tokutek has come out with a
stand-alone product I'm not aware of. At best you'll still need to deal with
Oracle to get the MySQL bits, at worse Tokutek's technology is as good as it
sounds and they are not an acquisition target for Oracle. (Assuming you
don't want to deal with Oracle, which you're reasoning above implies. Many
people won't care about that though).

>
> > I've googled around for a while, but all the benchmarks I've found
>
>> commit one or more "fatal flaws", which render the benchmark pointless
>> at best:
>>
>
> Doing real benchmarks is an involved, expensive process, and both DBMSes
> are fast-moving targets which need to be benchmarked every year. I've
> talked with Percona about doing some MySQL vs. PostgreSQL head-to-heads, but
> in the absence of specific funding that's unlikely to be completed.
> Besides, we both really want to do head-to-heads with MSSQL & Oracle, not
> each other.
>
>
Actually in this case, "real" benchmarks should be pretty easy. You already
have a running application that you have numbers on. I'd say swap in a copy
of xtradb, a copy of tokutek's tech, and a copy of Postgres (after all, it's
only "5 minutes" to change the queries around), and benchmark all three.
Realistically it would probably take a dedicated week to do it, but you're
probably making a 3 year decision, so this bit of time up front seems like a
good way to go. (And don't forget to publish your results afterwards).

Robert Treat
play: http://www.xzilla.net
work: http://www.omniti.com/is/hiring