On 1 February 2012 21:42, Oliver Lloyd <[email protected]> wrote: >>The disadvantage is that writing large numbers of samples to a >>database is likely to be slower than writing to a file. > > Agreed. In principle it is possible to tune inserts to high levels (we got > it up to 40k a second) but that was with fairly lightweight data and > involved some tweaking of the db params which is non-standard. > > I guess a better approach might be either to write to a db non synchronously > using a flat file as a buffer
That could be done outside JMeter by writing something to "tail" a CSV output file. [Or of course, wait until the test is finished and then do the upload]. > or maybe using some sort of mem cache. The > latter would offer the best throughput but with obvious risks. Indeed. >>What makes you think that Listeners depend on the output format? > > I was thinking in terms of loading results offline. Ideally if results are > written to a db then it should be possible to point a listener at this > source. We'll review the common code on that. As a workround, just use standard database tools to create a CSV file which JMeter can then load. But surely there are lots of tools for analysing database data directly - I'm not sure that should be a high priority. > Thanks for your replies; we'll try and spend some more time on this. I'm not saying it should not be attempted, just that there are potential downsides. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
