prediction.tex

\section{Query Resource Usage Prediction}
\subsection{Resource allocation characteristic in K8S}
It is always a hard task to predict the resources a query will consume in particular physical machine environments, because the result may different by blablablah.  There are already bunch of work of research[14,44] on how to model a prediction system of resource usage of queries in DBMS, the result is (need add). For  kubernetes native database, it's even more important for us to have a prediction model, once the resource configurations is determined by HAWQ, there is no way to adjust them during the runtime of the query. If the resource allocated is less than the query need, query may run much slower than it should be. If we allocate too much resources than the query really need, the resource will be always occupied during the query execution, there is no way the over-allocated resource can be shared by other queries in k8s environment. Thus, a reliable prediction solution is one of the key point for a k8s native database to run more efficiently.