Most modern in-memory database systems rely on multi-version concurrency control to support real-time data analysis without interfering with concurrent writes. However, this is not a good fit for heterogeneous workloads. We find that long version chains are the root cause of the throughput reduction. In this paper, we exploit a scalable version-aware data partitioning and placement approach for heterogeneous workloads that incorporates a suite of optimized techniques to significantly reduce the overheads incurred both during the initial placement and during version ingestion at runtime. The experiment results show that the proposed approach achieves 2x performance improvement compared with existing state-of-the-art approaches.