Hi Peter, Noticed a similar behavior while working on cTAKES REST module. The in-memory HSQL in my case was stressing the application server memory and ended up slowing down the process whereas mysql performed better. Also the engine you use in MySQL matters as well.
We did a testing on MySQL based UMLS dictionary using multiple pods running ctakes rest and it was scaling fairly well. But havent explored with 100+ connections. But i guess with connection pool configurations in MySQL DB it should be manageable. Hope it helps. On Thu, Feb 25, 2021 at 7:37 PM Peter Abramowitsch <pabramowit...@gmail.com> wrote: > Hi all, > > As an experiment I extracted my rather large HSQL UMLS dictionary into a > local MYSQL instance and ran the equivalent of 3 simultaneous ctakes > pipelines with the overlap lookup annotator against it with a set of 1000 > notes. > > Comparing that with the same setup running against the traditional > in-memory HSQL database (three separate instances), I was surprised to find > that the Mysql implementation it was 30% faster even though it is an out of > process DB > > Has that been anyone else's experience as well? And if so, do you have any > experience with a MYSQL based UMLS dictionary with 100+ pipeline > connections? > > Peter > -- Regards, Gandhi "The best way to find urself is to lose urself in the service of others !!!"
