Abstract: Internet 　applications 　create　massive　small　spatio－temporal 　data 　files 　in 　cloud 　environments．Therefore　a　method　aimin g　to 　raise 　the　processing　efficienc y　of 　small　files 　in　HDFSa　data　scheme combining　user 　access 　and　data 　featuresis 　proposed．This 　scheme 　re gards 　a　user 　access 　stream　as 　file re quest 　se quenceand　constructs 　a　characteristic 　se quence　by　spatio－temporal 　attribute 　extraction．A characteristic 　template 　of 　different 　user 　access 　patterns 　is 　formed　when　anal yzin g　the 　characteristic 　sequence　by　template　matching．Then　merger－related 　files 　are 　anal yzed．Experimental 　results 　show　that our　scheme 　improves 　the 　stora ge　efficienc y　for 　small　files and　also 　decreases　network　application 　response　times．