Dependency design is one of the key factors of the overall performance of a large application system for both runtime phase and development phase. The 2 features of cloud enabled large applications make this dependency design more critical. The first is that service is always distributed and data application is always partitioned. To be distributed and partitioned means there are more delay and complexity with the dependencies. The other feature of cloud is there are more failures in the cloud application systems with the commodity hardware than those enterprise systems with high-end professional equipments. One system failure may cause cascading failures and impact the stability and performance of the whole system, just like traffic jams caused by one traffic accident. This paper addresses a number of architectural patterns and solutions that have been found to build a large application especially for cloud system runtimes with well managed dependencies. The patterns also include our successful experiences to build a high throughput, low latency, and highly fault tolerant distributed service engine for cloud applications. © 2013 IEEE.