[geeks] An Interesting Idea

[Charles Shannon Hendrix]

Sridhar Ayengar wrote:
> Charles Shannon Hendrix wrote:
>> On Thu, 22 Feb 2007 16:14:26 -0500
>> Sridhar Ayengar <ploopster at gmail.com> wrote:
>>
>>> How about a bogosort benchmark?
>>>
>>> Just remove all semblance of science from benchmarking.
>> There already is a bogosort benchmark.  It was written by Oracle for the SQL
>> performance tests...
> 
> Do the results actually show anything interesting?

Of course they do: they always show Oracle winning.

In case you didn't get it: I was being a smartass.  Oracle wrote a lot 
of transaction database benchmarks, and of course they favor oracle 
rather heavily.

It's about as bogus as you can get.

Aside: something I always liked about some minicomputers and mainframes: 
you could actually find out how expensive various operations were in 
terms of instructions executed for the full path.

For example, the number of instructions to process a keystroke, write a 
block, to a transaction, or run an SQL query.

It helped you see performance killers in hidden layers and that sort of 
thing.

IBM used to do a lot of benchmarking like that to reduce overall load 
for large numbers of users, and to optimize latency of various 
interfaces, including things like how much time can we get away with 
before a user feels left out.