Scheduling Of IO Commands In Storage Networks

With the developing of the Internet, a dramatic growth of enterprise data storage capacity can be observed in the last couple of years. People require more and more performance, capacity and manageability of the storage with the time goes by. So the relative slow developing speed of the storage compared to the other parts of the computer system such as CPU, network bandwidth and so on are becoming the bottleneck of IT's progress. Now, the SAN (Storage Area Network) is a popular technology to solve these problems. It can ensure the reliability, serviceability, scalability and availability of the storage. But now it obviously can not satisfy the increasing need of market, so many researchers are working on improving the technology of SAN.In SAN, the mature network technology is used to substitute for the IO bus. Usually, the common switch is used in the storage network especially in the iSCSI storage system. Because the common switches are not aware of the environment it is used in, it can not make optimization for the storage network. In this dissertation, an optimization of storage network is presented.There are three main components in an SAN environment such as server systems, storage devices and interconnect devices. So methods to improve the SAN are mainly focused on these three components. Most of the researches are focused on the server systems or storage devices, but this paper focus on the IO commands transferred in the switch which is used to connect the servers and storage devices.In this paper, the progress of the IO commands transferred in the iSCSI protocol are considered particularly. The iSCSI PDUs transferred between the initiator and the target are divided in two kinds: one is the commands with the read or write data; the other is the commands related with the waiting for the resources, which are called short commands. These short commands passing the switch used in storage network are transferred in highest priority, in which aim to improve the performance and resource efficiency of the whole storage system. Theorical testify using the queueing theory showed that the optimization reduce the average delay time of the storage network. The results of the simulation tests showed that the optimization reduce the average disposal time, improve the throughput and the resource utilization of the storage network. In the testing experiment, the linux progress is modified. In order to transfer the commands with the priority, the position of the IO acknowledgment commands add to the commands list has been changed. Using the testing tool gained the results that the optimization can improve the average response time and the throughtput of the storage network.