On 28 Březen 2012, 17:44, Thom Brown wrote:
> On 28 March 2012 16:30, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> "Tomas Vondra" <tv@fuzzy.cz> writes:
>>> On 28 Březen 2012, 16:38, Gregg Jaskiewicz wrote:
>>>> They seem to claim up to 70% speed gain.
>>>> Did anyone proved it, tested it - with PostgreSQL in particular ?
>>
>>> I really don't expect such difference just due to switching to a
>>> different
>>> kernel. There's a space for infinite number of tweaks there (using a
>>> different default fs parameters, adding better support for the new
>>> Niagara
>>> T4 CPU not available to RedHat yet etc.).
>>
>> AFAIK, Oracle Linux is still just rebranded RHEL, with some very minimal
>> amount of additional engineering effort put in. It's not likely that
>> they are so much smarter than everybody else who works on Linux that
>> they can find huge across-the-board speedups that nobody else has found.
>
> Reminds me of when Oracle claimed a 70x speed increase in MySQL
> cluster ("Delivers up to 70x More Performance for Complex Queries"),
> and the ability to process a billion queries per minute. Upon closer
> inspection, the tables used in the "billion tables" benchmark were all
> in-memory tables with no joins and distributed across 8 servers. And
> the increases over the previous version weren't fair either because
> the tests were using different hardware *and* one of them was
> virtualised. They also didn't appear to want to disclose any further
> details of the hardware differences.
>
> So basically setting up unrealistic scenarios to get the highest
> hype-making numbers, and ensuring the important context of those
> numbers is in the footnotes somewhere.
Not to mention they actually prohibit independent benchmarking in their
license. They claim they do that to prevent "unfair" benchmarks executed
by people lacking the necessary knowledge, but considering the benchmarks
they publish ...
Tomas