BDE alternatives: Who's faster?

         This article presents the result of speed characteristics investigations; BDE alternatives are compared to Borland Paradox with BDE. Diagram 1 presents the results of all existing speed tests for the products mentioned above. Diagram 2 presents reading speed test results, and Diagram 3 presents writing tests results. Comparison of all these products by searching speed when searching for all the records matching the specified search criteria, is presented at Diagram 4 (find first and find next results are compared for equality on 5 conditions. They are: Integer, Sting, Date/Time, Integer & String & DateTime, String with partial compare). Diagram 5 shows searching speed when searching for the single record matching the search criteria of the 'equiality' type (Locate by the same five conditions). Diagram 6 presents the result of comparison of the same DB engines by filtering speed (the same five conditions). Diagram 7 shows the results of comparison by BLOB fields processing speed.

Table 1. Tests passing time, sec.

         Well, let us now review it all.
         1. EasyTable by AidAim Software is an undisputable leader in searching and I/O operations speed.
         2. Paradox + BDE (native driver, supplied with the delivery set) by Borland Software Corporation shows quite good results in speed. This combination even exceeds a bit EasyTable, our tests leader, in data writing speed; but even if writing operations sort with reading operations as 1:1 (as for real applications, this ratio may be much greater), EasyTable still wins over Paradox in I/O operations speed (it is almost 3 times faster). Here we must say, that this is the only system of all the considered, which uses BDE; this feature sometimes becomes inconvenient due to the large size of the compiled code. All the other tested systems are 5 to 10 times more compact.
         3. Pity, but such a well-known and widely advertised in the Internet as a high-speed DB engine system as DBISAM did not show high results in speed and lost before the leader in the sum of all the tests, having appeared over 10 times slower then EasyTable. In the input-output time it remains behind the EasyTable, being almost 100 times slower! Moreover, DBISAM is equipped with a very slow searching engine. In the Find category EasyTable, our leader, is over 20 times faster than DBISAM. We should note that Find next is executed especially slowly; for example, in complex inquiry (Find next all fields) EasyTable is almost 500 (!!!) times faster. DBISAM is almost 100 times slower than EasyTable in Locate and over 200 times - in filtering. Hence, to our opinion the testing results cited by DBISAM as a confirmation of their high results in speed, are not quite correct, for the testing principles are far from being perfect, as well as the way to produce the result (I/O times are not measured; single operation time being measured is comparable to the measurements precision itself; only special operations are measured; the source code given is beyond compilation and so it may not be used for testing ), the products assortment for testing is very tendentious.
         4. The slowest of all DB engines that passed our tests successfully is MiniTable. EasyTable is 176 times faster than it is! It looks like this product's perfection finally stopped over a year ago.
         5. Obviously bad results shown by TurboDB became an unpleasant surprise. In I/O operations EasyTable is over 150 times faster than TurboDB is. Moreover, this DB engine appeared unable to pass our testings due to errors that arosed with testing searching (Locate) and filtering (due to bugs with strings and dates). TurboDB doesn't have Find first / Find next featuers at all. Even in those rare cases when we managed to measure its searching and filtering characteristics, TurboDB proved to be the most slow system: EasyTable is 10,000 (!!!) times faster in searching by Locate and almost 500 times faster in filtering. So, tests had really shown that TurboDB isn't worse its "quick" name - let this trademark never mislead anyone else!
         6. TinyDB also appeared unable to pass testings due to errors in testing Find and filtering. Moreover, like TurboDB it showed very low speed as in the I/O operations, so as in the searching engine testing, too.
         7. We had also tested some other systems, but their results proved to be even worse than those of this review's leaders; over this, we found some of those products worthless mentioning in this article due to some of their features.
         To sum up, we must say that EasyTable proved to be the fastest to date DB engine. TurboDB and TinyDB are not recommended to use due to multiple errors in their searching algorithms. Moreover, these two systems seem to be absolutely hopeless in achieving high performance results.
         Note. When reviewing the summary productivity of the tested DB engines we supposed that all the operations take the same part in the total system's productivity; surely, you'll never meet such a situation in real applications. Nevertheless, having received the above-mentioned results and having analysed the real frequency of using this or that operation in your application on the example of a typical task, you may learn, which of the systems listed above will be faster, and how much faster, for your particular case. However, basing on the results given in the present article, EasyTable by AidAim Software will probably become the fastest system in your case, too.

         We give source code of the testing program for each DB engine. You may also download compiled testing programs with sources. Due to some distinctions of functions realization in different programs, we used different testing programs for any of the products being tested, using the same measurement principles. Any comer may carry out the same measurements and make sure in reliability of the results presented here. You may also be free in testing the products presented here and change the measurements principles in order to obtain, for example, the dependency of DB engine(s) operating speed on the number of records. Furthermore, you may change any of the programs presented here in order to test any other product. In any case, with any publishing of any investigations using the materials presented here (including program(s) for measurements) or the measurement principles, you must needs refer to the author and the article itself.

Raymond Mortimer.