I/O Performance Optimization For Typical Applications In Cloud Environment

Cloud computing is one of the hottest topic in research and industry. Its applications require large scale service and are usually data-intensive. The capability of current plat-form, including software and hardware, does not match the features of these applications. They result in high energy consumption and low efficiency of the current system architec-ture and are challenging the design of the system architecture. With vast amount of data, the performance of I/O subsystem become more and more critical to cloud computing.Optimization to I/O performance is deserved and can be widely applied to many applications. By application driven, streaming media server, email server, and Web server are chosen as the representatives for concrete research. To uncover the problem, we analyze these specified applications on the sides of the hard disk, the network, and the memory access to pursue performance.Interactions of client sometimes hurt the quality of service provided by. a streaming media server for insufficient I/O capacity. Frames of a streaming media are clustered into two types of block according to their type, I or non-I, and are stored in an order determined by its timestamp to accommodate highly interactions by exploiting the in-trinsic feature of the media themselves. A caching algorithm based on interval caching is designed to against the challenges of interactions. Competitive analysis is applied to evaluate the impact of frame reorganization to caching algorithm.With a demonstration of several scenarios of email usage, we disclose that there are some redundancy among all users’data. The architecture of an email server is modified for data deduplication, which is leveraged to save space and to reduce bandwidth of disks on server. Furthermore, data deduplication can also help reduce network bandwidth by expanding SMTP protocol to introduce a mechanism to search before transfer.Finally, data of these applications are classified into service data and meta data. A smart memory system is designed to improve efficiency of data accessing including a memory controller, a small computation chip, a protocol, and a network interconnecting them. Data are stored into the memory device, including the smart memory, SRAM, and DDRx memory, which matches its own accessing feature to data accessing feature.