Distributed　data　stream　processing　systems(DSPS)　are　widely　used　in　real-time　massive　data　processing　scenarios　for　its　characteristics　of　real-time　and　high　throughout.　In　the　real-world　DSPS,　the　fluctuating　arrival　rate　of　the　input　data　leads　the　consuming　computing　resource　of　DSPS　to　be　time-variable.　To　guarantee　the　performance　of　DSPS,　the　accurate　prediction　of　DSPS＇s　consuming　resources　is　necessary.　In　this　paper,　we　proposed　approaches　to　make　the　online　prediction　of　computing　resources　that　DSPS　consumes.　We　monitor　the　usage　of　computing　resources　such　as　CPU　and　memory　in　a　DSPS,　and　use　temporal　data　streams　clustering　algorithm　and　linear　regression　method　to　make　online　prediction　of　CPU　resources　and　memory　resources　respectively.　Our　prediction　approaches　are　proved　efficient　and　quickly　enough.