The key-value storage based on LSM-tree converts random writes to sequential writes is widely used in big data storage, but its frequent compaction operation brings write amplification problem which is not friendly to SSD. Key-value separation can effectively reduce write amplification, but key-value separation structures often exhibit poor range query performance and intensive update performance. In this paper, we propose SplitKV, which improves compaction operation to split operation based on key-value separation, and divides keyranges according to data distribution. Improves range query performance by converting a large-scale operation into multiple small-scale operations, and greatly enhances its ability to read missing key. We compared SplitKV with classical key-value storage systems, and experiments showed that SplitKV improved range query performance by up to 211% and readmissing performance by 51% compared with RocksDB.