Si tiene giovedì 22/04 dalle 14 alle 16 il seminario aperto a tutti su Teams (link per accedere) dal titolo “Manual Performance Testing and Automatic Performance Tuning” organizzato dalla prof. Vittoria De Nitto Personè, docente di Performance Modeling of Computer Systems and Networks nei corsi di laurea magistrale di Ingegneria Informatica e di Ingegneria dell’Automazione.
Animano la lezione ‘speciale’ Stefano Cereda e Giovanni Paolo Gibilisco di MOVIRI, consulting & technology company, fondata nel 2000 come start up del Politecnico di Milano e presente in diverse parti del mondo (Milano sede principale, Boston, Los Angeles, Singapore), che affrontano il tema delle applicazioni di tecniche di performance a sistemi reali e complessi, dei loro limiti e innovazioni possibili.
Maximising the performance of IT systems is critical to reduce hardware and software costs. Each change to the system has to be thoroughly validated to avoid disrupting the performance.
However, traditional performance testing is a tedious process that involves several manual steps such as running experiments, collecting measurements and creating stochastic models that allow to make predictions about the performance of a system. Modern IT systems are becoming more complex, to the point where, when performance problems arise, traditional modelling gives no helpful insight to the human expert.
On the other hand, automatic performance tuning automates the entire performance testing process and replaces the human with machine learning models that can quickly and reliably improve performance.
In this talk, we start by covering traditional performance testing techniques and apply them to the MongoDB DBMS, showing how they can be used to make accurate predictions about actual data. Then, we show how a quick reconfiguration of the Operating System considerably improves performance. Finally, we show how machine learning can automatically handle the reconfiguration process.
(A fundamental familiarity with Queuing theory (Little’s law) is required to follow the first section, but no ML-related knowledge is required.)