iSCSI storage system is the most widely used network storage system, and soft real-time services are the most common type of business. However, the traditional I/O scheduling algorithms can't work very well for iSCSI storage system oriented soft real-time services. In this paper, the mathematical model of I/O scheduling in iSCSI storage system for soft real-time services is analyzed and discussed, and it proves that this is a multi-objective optimization problem. HB-SCHED is proposed in this paper, which is a heuristic I/O scheduling algorithm for soft real-time services oriented iSCSI storage system. The simulation results show that HB-SCHED not only has better quality in terms of real time than SCAN and can gain higher disk throughput than EDF, but also can adjust the weights between real time and disk throughput by changing the parameter value. HB-SCHED can be well adopted in soft real-time services oriented iSCSI storage system.

No abstract

Apache is the most widely used HTTP server in the world. Requests per second, transfer rate, time per request is the widely used indices in benchmarking HTTP server. This paper used benchmark test program ab which is widely used compared the performances of Apache between the multi-core network processor OCTEON5860 and XEON E5506. The results show that the performance of OCTEON 5860 is better than XEON E5506. Finally, through analyzing the advantage of network processor OCTEON5860 explained the benchmarking result.

No abstract

When Running Linux on a multi-core processor, all of the network processing is processed by the operating system and the processing efficiency is very poor. To solve this problem, we describe a Multi-core Network Processing System (MNPS) in this paper. The MNPS divides the cores of a multi-core processor into two groups. One group runs MNPS-Linux, which is a modified Linux only retaining the same socket interface to the user applications. The other group runs MNPS-Stack, which processes the network stack without operating system and directly operates the hardware resources. Because of the offloading of network processing and no interrupt in MNPS-Linux and the directly operating the hardware resources and no operating system in MNPS-Stack, MNPS is very efficient. We can see from the result of the experiment that MNPS can get high performance but consumes less CPU resource.