Обсуждение: DB benchmark and pg config file help

Hello List,

Not sure to which list I should post (gray lines, and all that), so
point me in the right direction if'n it's a problem.

I am in the process of learning some of the art/science of benchmarking.
  Given novnov's recent post about the comparison of MS SQL vs
PostgresQL, I felt it time to do a benchmark comparison of sorts for
myself . . . more for me and the benchmark learning process than the
DB's, but I'm interested in DB's in general, so it's a good fit.  (If I
find anything interesting/new, I will of course share the results.)

Given that, I don't know what I'm doing.  :|  It seems initially that to
do it properly, I have to pick some sort of focus.  In other words,
shall I benchmark from a standpoint of ACID compliance?  Shall I
benchmark with functionality in mind?  Ease of use/setup?  Speed?  The
latter seems to be done most widely/often, so I suspect it's the easiest
standpoint from which to work.  Thus, for my initial foray into
benchmarking, I'll probably start there.  (Unless of course, in any of
your wisdom, you can point me in a better direction.)

 From my less-than-one-month-of-Postgres-list-lurking, I think I need to
be aware of at /least/ these items for my benchmarks (in no particular
order):

* overall speed (obvious)

* mitigating factors
   - DB fits entirely in memory or not (page faults)
   - DB size
   - DB versions

* DB non-SELECT performance.  A common point I see in comparisons of
   MySQL and PostgresQL is that MySQL is much faster.  However, I rarely
   see anything other than comparison of SELECT.

* Query complexity (e.g. criteria, {,inner,outer}-joins)
    ex.    SELECT * FROM aTable; vs
    SELECT
      FUNC( var ),
      ...
     FROM
       tables
     WHERE
         x IN (<list>)
    OR y BETWEEN
          a
      AND b ...

* Queries against tables/columns of varying data types.  (BOOLEAN,
   SMALLINT, TEXT, VARCHAR, etc.)

* Queries against tables with/out constraints

* Queries against tables with/out triggers {post,pre}-{non,}SELECT

* Transactions

* Individual and common functions (common use, not necessarily common
   name, e.g. SUBSTRING/SUBSTR, MAX, COUNT, ORDER BY w/{,o} LIMIT).

* Performance under load (e.g. 1, 10, 100 concurrent users),
   - need to delineate how DB's handle concurrent queries against the
     same tuples AND against different tuples/tables.

* Access method (e.g. Thru C libs, via PHP/Postgres libs, apache/web,
   command line and stdin scripts)

# I don't currently have access to a RAID setup, so this will all have
   to be on single hard drive for now.  Perhaps later I can procure more
   hardware/situations with which to test.

Clearly, this is only a small portion of what I should be aware when I'm
benchmarking different DB's in terms of speed/performance, and already
it's feeling daunting.  Feel free to add any/all items about which I'm
not thinking.

The other thing: as I'm still a bit of a noob, all my use of the
Postgres DB has been -- for the most part -- with the stock
configuration.  Since I'm planning to run these tests on the same
hardware, I can pseudo-rule out hardware-related differences in the
results.  However, I'm hoping that I can give my stats/assumptions to
the list and someone would give me a configuration file that would /most
likely/ be best?  I can search the documentation/archives, but I'm
hoping to get head start and tweak from there.

Any and all advice would be /much/ appreciated!

Kevin

On 1/17/07, Kevin Hunter <hunteke@earlham.edu> wrote:
> Hello List,
>
> Not sure to which list I should post (gray lines, and all that), so
> point me in the right direction if'n it's a problem.
>
> I am in the process of learning some of the art/science of benchmarking.
>   Given novnov's recent post about the comparison of MS SQL vs
> PostgresQL, I felt it time to do a benchmark comparison of sorts for
> myself . . . more for me and the benchmark learning process than the
> DB's, but I'm interested in DB's in general, so it's a good fit.  (If I
> find anything interesting/new, I will of course share the results.)

Just remember that all the major commercial databases have
anti-benchmark clauses in their license agreements.  So, if you decide
to publish your results (especially in a formal benchmark), you can't
mention the big boys by name. [yes this is cowardice]

merlin

On 19 Jan 2007 at 8:45a -0500, Merlin Moncure wrote:
> On 1/17/07, Kevin Hunter [hunteke∈earlham.edu] wrote:
>> I am in the process of learning some of the art/science of benchmarking.
>> Given novnov's recent post about the comparison of MS SQL vs
>> PostgresQL, I felt it time to do a benchmark comparison of sorts for
>> myself . . . more for me and the benchmark learning process than the
>> DB's, but I'm interested in DB's in general, so it's a good fit.  (If I
>> find anything interesting/new, I will of course share the results.)
>
> Just remember that all the major commercial databases have
> anti-benchmark clauses in their license agreements.  So, if you decide
> to publish your results (especially in a formal benchmark), you can't
> mention the big boys by name. [yes this is cowardice]

"Anti-benchmark clauses in the license agreements"?!?  Cowardice indeed!
  <wry_look>So, by implication, I should do my benchmarking with
"borrowed" copies, right?  No sale, no agreement . . . </wry_look>

Seriously though, that would have bitten me.  Thank you, I did not know
that.  Does that mean that I can't publish the results outside of my
work/research/personal unit at all?  Or do I just need to obscure about
which DB I'm talking?  (Like Vendor {1,2,3,...} Product).

Appreciatively,

Kevin

On Fri, Jan 19, 2007 at 09:05:35 -0500,
  Kevin Hunter <hunteke@earlham.edu> wrote:
>
> Seriously though, that would have bitten me.  Thank you, I did not know
> that.  Does that mean that I can't publish the results outside of my
> work/research/personal unit at all?  Or do I just need to obscure about
> which DB I'm talking?  (Like Vendor {1,2,3,...} Product).

Check with your lawyer. Depending on where you are, those clauses may not even
be valid.

On 19 Jan 2007 at 10:56a -0600, Bruno Wolff III wrote:
> On Fri, Jan 19, 2007 at 09:05:35 -0500,
>   Kevin Hunter <hunteke@earlham.edu> wrote:
>> Seriously though, that would have bitten me.  Thank you, I did not know
>> that.  Does that mean that I can't publish the results outside of my
>> work/research/personal unit at all?  Or do I just need to obscure about
>> which DB I'm talking?  (Like Vendor {1,2,3,...} Product).
>
> Check with your lawyer. Depending on where you are, those clauses may not even
> be valid.

<grins />

/me = student => no money . . . lawyer?  You /are/ my lawyers.  ;)

Well, sounds like America's legal system/red tape will at least slow my
efforts against the non-open source DBs, until I get a chance to find
out for sure.

I really do appreciate the warnings/heads ups.

Kevin

BTW: I'm currently located in Richmond, IN, USA.  A pin for someone's
map. :)