CityGML　provides　an　effective　approach　to　the　sharing　and　inoperability　of　3D　GIS　spatial　data.　However,　due　to　the　large　amount　of　data,　CityGML＇s　resolving,　transferring,　and　rendering　processes　during　visualization　are　inefficient　in　a　standalone　system,　causing　considerable　delays　in　viewing　the　information　of　large　3D　maps　for　users.　The　cloud　computing　technique　is　introduced　to　address　these　problems　in　this　paper.　The　Hadoop　Distributed　File　System　(HDFS)　is　employed　to　store　the　huge　amount　of　CityGML　data.　A　MapReduce-based　parallel　CityGML　data　visualization　scheme　is　proposed.　A　Hadoop-based　public　cloud　for　a　3D　city　information　service　is　constructed　in　the　cloud,　allowing　cloud　users　to　interact　with　the　cloud　via　the　service　interface　and　obtain　the　desired　high-quality　3D　city　scenarios.　The　visualization　effectiveness　and　interoperability　efficiency　of　the　large-scale　CityGML　3D　virtual　city　data　are　improved.