The rapid growth of blockchain systems such as Bitcoin and Ethereum has led to an escalating storage burden on nodes. The storage scalability of blockchain has become an urgent problem. On-chain storage approaches introduce cross-shard transaction coordination and index maintenance management problems. Hybrid on-chain/off-chain storage approaches execute transactions off-chain, which face reliance on third-party and data consistency problems. To solve the above problems, we propose a novel storage-compute separation blockchain architecture. Computations related to transactions and consensus are still processed on-chain by blockchain nodes. While the block data is stored in distributed file systems shared by blockchain nodes. Meanwhile, to ensure data consistency, integrity and read/write efficiency, we design a new read and write workflow based on four modules for the proposed architecture. We implement the proposed architecture based on Hyperledger Fabric. Experimental results for three scenarios (transfers, world-state queries, and historical data queries) show that compared to Hyperledger Fabric, our architecture reduces storage by 49.9% with only TPS loss of 3.29% and latency loss of 9.20% on average.