Relabling to provide focus on this thread the KVM/FreeBSD specific issues - I'll deal with that the main thread.
So, if I could put forward to you a few suggestions. 1) Review the existing tests (which you have done somewhat already) 2) Define a clear and relevant intent for benchmarking SQLite 3) Work with Michael (from Phoronix) and myself to codify this intent into a set of test cases and test suites I'll let this fork of the thread settle for a bit before taking it off list. Regards, Matthew -------- Original Message -------- Subject: Re: [sqlite] SQLite behaviour on FreeBSD and KVM From: Simon Slavin <slav...@hearsay.demon.co.uk> To: General Discussion of SQLite Database <sqlite-users@sqlite.org> Date: 09/29/2009 08:40 PM > On 29 Sep 2009, at 10:29pm, Matthew Tippett wrote: > >> If there is anyone who is interested in assisting in improving the >> quality/value/functional interest of the benchmarks, then please >> advise. > > In SQLite, when you know you are making many changes and don't need to > consult the data until you're finished, you surround the changes with > BEGIN TRANSACTION and END TRANSACTION. This makes them into one big > update rather than lots of little ones, and it means that disk gets > updated just once (handwave here) rather than after each command. > Naturally, this is hugely faster. No way should 2500 inserts in > SQLite take 14 minutes. > > So much faster that, as Pavel noted upthread, it's suspiciously like > what you're seeing in the result for KVM. This suggests that KVM is > not really writing results to disk immediately. Putting those INSERTs > into one transaction could make Ubuntu 9.10 (not KVM) as fast or > faster than the result you're getting for KVM. Sorry, I have no Linux > to test it on. > > You could modify the sqlite test to reflect this. You could perhaps > turn sqlite-2500-insertions.txt into 50 transactions, with BEGIN > TRANSACTION and END TRANSACTION around each 50 INSERT commands. Or > you could have two tests: make two copies of sqlite-2500- > insertions.txt, leave one as it is and put BEGIN TRANSACTION and END > TRANSACTION at the beginning and end of the other. This would test > both 2500 individual INSERT commands and one transaction of 2500 > INSERTs, testing SQLite as both kinds of application would need to use > it. > > I know nothing about KVM but I assume that it's operating correctly > here: since the entire machine is virtualised it doesn't matter that > it's not really writing to real disk. > > Simon. > _______________________________________________ > sqlite-users mailing list > sqlite-users@sqlite.org > http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
