On Mon, Feb 10, 2003 at 11:58:31AM +0000, Tim Sweetman wrote: > Andy, > > Andy Wardley wrote: > > > > Penny Bamborough wrote: > > > We did write a > > > perl version of the streetmap engine (with some help from some very nice > > > people I might add) however performance tests on the system indicated that > > > the processing power required would cripple the server and our site would > > > be very short lived. > > > > [...] > > > > > The site has grown considerably since that time, we do use Win2k servers > > > with our own IIS extensions written in C++ to power the site > > > > I'm a little surprised by that. Although I must admit that I've never > > written IIS extensions in C++, I'm surprised that it offers significantly > > better performance than a mod_perl solution. > > I hope Penny has enough relevant experience to realise that Andy's > performance discussion above is predicated on other things being equal, > and that they quite likely aren't.
Are there any definitive (FSVO) benchmarks on different database abstraction systems? ADO vs ODBC vs DBI vs DBD vs ... ? Is there any advantage in using win32 from the POV of speedy database connections? Are the win32 named pipes systems fast? > > FWIW, a project I've worked on recently, where I've indulg^H^H^H^H^H^H^H > been forced into serious database renormalisation, has resulted in > queries that used to take ~20sec taking a second, or less... That strikes me as a piece of poor work by the original DBA. Is there any legal recourse along the lines of "you did a terrible piece of work. we paid you to do it. we had to get another contractor in to pick up the pieces, so we're going to bill you for his time" ? > > There are some reasonable good practices for performance (judicious use > of mod_perl, and the where-practicable use of decent programming > practice, database normalisation etc), but there's no substitute for > testing and investigating performance based on how the whole system will > operate. If the database isn't normalised, it's arguably just a set of flat files. Not normalising properly reduces the benefits of relational dbms /massively/. The main thing that the BIG systems implementors (logica spring to mind) can do which small privateer outfits can't is to employ people for the express purpose of stresstesting systems. I've written systems to load-test servers, but that won't spot the UI bugs... /joel
