Currently,　with　the　widespread　use　of　distributed　information,　traditional　clustering　algorithms　can　no　longer　meet　the　processing　needs　of　massive　information　both　in　terms　of　accuracy　and　computational　efficiency,　so　clustering　algorithms　based　on　the　Spark　distributed　platform　have　become　today＇s　research　hotspots.　The　K-Means　algorithm　can　be　widely　used　in　both　academic　research　and　business　by　virtue　of　its　ease　of　implementation　and　high　scalability.　However,　the　traditional　K-Means　is　based　on　Euclidean　distance,　which　is　not　applicable　in　some　scenarios.　And　it　is　inefficient　when　dealing　with　large-scale　data.　This　study　implements　and　tests　K-means,　K-means++,　Canopy+　K-means　algorithms　based　on　Euclidean　distance　and　Manhattan　distance　in　Spark　distributed　platform　and　analyses　the　performance　changes.　The　experimental　results　show　that　the　introduction　of　Manhattan　distance　makes　the　clustering　time　longer　and　the　optimisation　effect　of　different　algorithms　changes　differently.　©　2024　IEEE.