As heterogeneous CPU clusters become more prevalent in cloud environments, cloud service providers face the challenge of efficiently scheduling diverse services to harness the potential of CPUs with varying generations and architectures. We present HeteroBridge, a scheduling system that bridges service diversity and CPU heterogeneity through program similarity analysis. To address service diversity, we propose a novel program similarity analysis method that combines embeddings of control flow graphs (CFGs), library function usages, and global features, enabling the system to identify similar programs for resource demands estimation. To address CPU heterogeneity, we introduce the Relative Performance Ratio (RPR) to derive a unified capacity metric across different CPUs. By integrating program similarity analysis and unified capacity modeling, HeteroBridge assigns services to optimal computing nodes. We integrate HeteroBridge with Kubernetes (K8s) and evaluate its performance in heterogeneous environments. Experimental results demonstrate that the Mean Reciprocal Rank (MRR) of program similarity analysis improves by up to 2.37 \(\times \) compared to baseline methods. Simultaneously, service latency is reduced by up to 52% compared to Kubernetes’ default scheduler. These findings highlight HeteroBridge’s ability to deliver efficient service scheduling in complex cloud environments.