Re: ANN: kbmMemTable v. 4.00 released! And comments on the AidAimbenchmark.


2003-07-29 04:17:21 AM
cppbuilder81
Hi,
Quote
May be. But what kbmMemTable guys has published are not tests, it is the pack
of tricks just look at their source code. They did:
- Modified test program to achive a better results. What was the problem? It
looks like they can not win by measuring the time of the single operation -
instead of this they offer indexes re-creating once per 100000 operations.
With words "real life". Funny.
You throw around you with allegations.
That situation may not be real life to you, but it sure as h... is real life to lots and lots of kbmMemTable users.
As I have explained lots of times - but which you seem not to be willing to read, or even less understand - is
that the benchmark is not C4D's invention, its AidAim's. They chose to setup the benchmark as they did.
C4D chose to take the benchmark again with different number of records, equalling the string comparison routines.
Even in that test, kbmMemTable is much faster than SQLMemTable for less than 40.000 records.
For more than 40.000 records, the internal structure of SQLMemTable starts to pay off with on the fly indexes.
That has also been shown in the C4D test. The C4D test has _not_ hidden any information, and is testing on equal terms
with SQLMemTable. If you have some properties you can set on the SQLMemTable to make it perform even better for that
test suite, then by all means use it and publish the result along with information about what the change was.
Quote
- Modified their products for the better works on such "tests". They used
non-standard features of their product. I think, guys from Aidaim could do
the same but they did not. May be, tests war will be continued? New version
with some special tricks in the product code to win the test. Originally.
AidAim have maybe not employed that feature in SQLMemTable because SQLMemTable internally use b-tree structures and
would not benefit tremendously from batched index updates. The amount of work needed is in the same ballpark as doing
the updates on the fly.
kbmMemTable can benefit from batched index updates because it allows a much more efficient access to the TList
structures.
Quote

But we should take into account that we have no any guarantees that the
product will be better in real life - it was developed to win a test.
Which is exactly what C4D felt about the AidAim benchmark in the first place.
Quote
Well, let give us the arithmetic average time af each database operation, and
we will be able to make a coplex estimation for our developer's needs as we
like (approximate, of cource).
Aidaim's tests do this, but Components4Developers shows their virtual
"real-life".
What a bunch of rubbish.
Both tests show how long time an enduser application employing those specific test algorithms would take, nothing more,
nothing less. The real life result is what the end user is seeing.
best regards
Kim Madsen
XXXX@XXXXX.COM
www.components4developers.com
The best components for the best developers
kbmMW - kbmMemTable - kbmWABD - kbmX10