I agree with Wolfgang... My first version of the rules regarding expiration of events in a length sliding windows took 120s in my benchmark. After thinking a little bit, the second version took 2s. Changed dramatically the behavior of the application, now can process several thousands of events in no time! If I were you, my first try was to optimize the rules, doing some benchmark/profiling, to know where the bottleneck is. Just my 2c, Vítor Rui Moreira On 19/05/2014, 18:31, Wolfgang Laun wrote: -------------------------------------------------------------------------------- Declaração: A informação contida nesta mensagem, e os ficheiros anexos, é privilegiada e confidencial, destinando-se exclusivamente ao(s) destinatário(s).Se não é o destinatário (ou o responsável pela sua entrega ao destinatário) e recebeu a mesma por engano, fica notificado que é estritamente proibido reproduzir, guardar ou distribuir toda ou qualquer parte desta mensagem e ficheiros anexos. Por favor reencaminhe a mensagem para o responsável pelo seu envio ou contacte-nos por telefone e elimine a mensagem e ficheiros anexos do seu computador, sem os reproduzir. Disclaimer: The information contained in this message, and any files attached, is privileged and confidential, and intended exclusively for the included addresses. If you are not the intended recipient (or the person responsible for delivering to the intended recipient) and received this message by mistake, be aware that copy, storage, distribution or any other use of all or part of this message and the files attached is strictly prohibited. Please notify the sender by reply e-mail or contact us by telephone and delete this message and the files attached, without retaining a copy. -------------------------------------------------------------------------------- _______________________________________________

About rule structure and / or, i thought that builded graph was in conjunctive normal form to avoid this problem!? Should we use conjunctive normal form as is recommended into tho old CLIPS ? Le lundi 19 mai 2014 à 19:31 +0200, Wolfgang Laun a écrit :

Yes, we have done some benchmarking and profiling. And we have found some improvements. We are processing a lot of data. We receive price changes from foreign exchange markets, and have customers who have automated trading strategies using this data as facts in individual drools sessions. Ok, we will keep looking to optimize, however I am doubtful that we will see much improvement. -- View this message in context: http://drools.46999.n3.nabble.com/rules-users-Drools-and-GPU-tp4029607p40... Sent from the Drools: User forum mailing list archive at Nabble.com.