Re: Linux betas NT in TPC testing, running Oracle8

In article <7fnh52$q4v$1_at_nnrp02.primenet.com>,   Stephen Edwards <ja207030_at_primenet.com> wrote:
> Bill Chess <chess_at_la.com> wrote:
>
> : Here's the story:
> : http://rpmfind.net/veillard/oracle/

This now points to a legal notice pointing out that they can no longer publish the results. Too bad nobody glammed on to the bottom line numbers. One of the reasons that there is such strict control of TPC numbers is that vendors don't want $/TPC benchmarks done on system like FreeBSD or Linux.

> Hmmm... obviously a "Linux" site.
>
> : It's sponsored by more than one company (unlike the Mindcraft study),
> : but I have to wonder if maybe Oracle deliberately cripples their NT
> : version somehow.

Actually, Oracle is just a bit more paranoid than SQL Server. In the UNIX world, one uses process forks to handle multiple TCP connections. In the NT world, the traditional tactic is to use threads because the context switching is so burdensome. Unfortunately, when you have hundreds of threads running in a single process, you run a high risk of corrupting memory in very nasty ways. The best cure is to put Mutex locks between routines that modify common memory. Since library calls could also modify common memory, the best approach is to put mutex locks in front of nearly everything. The bad news is that a heavily loaded system (TPC benchmarks deliberately generate a heavy load), the system can go into gridlock. You have threads waiting for threads that are waiting for threads. It gets really ugly.

> The test is completely bogus. But I'm not suprised in the slightest.
>
> "No specific operating system tuning was done, all the default NT services
> were running, with the default priorities and system parameters."

Yes, I got a chuckle out of that too. Most UNIX systems by definition have to run certain defauld processes to function effectively. On NT, a database must run on a system that has been "stripped to the wheels". Again this is due to the context switching characteristics. If the NT system has do do it's own DNS lookups for it's clients, you now have two context switches per connection. Then you have additional overhead that is not part of the "pure" benchmark tests.

Unfortunately, the tests didn't make it to Germany, where such "conspiracy of silence" is considered a form of fraud. There are many interesting benchmarks available to compare Linux and NT. If you can read German, the results are startling.

> Oh no... not biased in the slightest.</SARCASM>

Actually, it's amusing to see how many tweaks Microsoft has to do to get a good benchmarke out of NT. Even then, it's interesting to see the costs of the systems. When the include the costs of the Client Access licenses, the prices are amazing.

> [] "No footnote for you!" -- Footnote Nazi
> --
> .-----.
> |[_] :| Stephen S. Edwards II | http://www.primenet.com/~rakmount
> | = :| "But something's wrong. It takes me a moment to pin it down;
> | | the monitor's different. Instead of the nice 17' Trinitron,
> |_..._| there's a 15' raster gun in a dirty plastic case." -- Ben in ASR
>

--
Rex Ballard - Open Source Advocate, Internet Architect, MIS Director
http://www.open4success.com

-----------== Posted via Deja News, The Discussion Network ==----------
http://www.dejanews.com/       Search, Read, Discuss, or Start Your Own