Hadoop　distributed　file　system　(HDFS)　has　been　widely　used　in　various　clusters　to　build　large　scale　and　high　performance　systems.　However,　it　is　designed　to　mainly　handle　big　size　files,　therefore　the　performance　processing　massive　small　files　is　relatively　low　because　of　huge　numbers　of　small　files　imposing　heavy　burden　on　Namenode　of　HDFS.　Focusing　the　problem　about　HDFS　when　processing　small　files,　an　approach　to　improve　I/O　performance　of　small　files　on　HDFS　is　introduced.　Our　main　idea　is　to　merge　small　files　in　the　same　directory　into　large　one　and　accordingly　build　index　for　each　small　file　to　enhance　storage　efficiency　of　small　files　and　reduce　burden　on　Namenode　caused　by　metadata.　Furthermore,　a　kind　of　Cache　strategy　to　improve　the　reading　efficiency　of　small　files　on　HDFS　is　presented.　Relevant　design,　data　structure　and　implementation　are　described.　The　experimental　results　indicate　that　the　method　proposed　can　improve　the　efficiency　of　processing　massive　small　files　on　HDFS.