Linux Network Throughput Control

Network throughput control is meaningful in multiple thread server for ensuring the system overload lived in a acceptable range and every processes can get bandwidth equally.This essay states the design principle and implementary method of throughput control in linux system. We implement it by intercepting the system call. This program was loaded as kernel module, taking charge of interception of system calls by modifing syscall table, access mechanisim would decide whether to execute current process according to the security policy, to guarantee no policy voialation. In design of network throughput control, we abide by follow disciplines: (1) reliable. Enforce throughput control to all processes, no exception. (2) transparency. We don't need to modify any programs that exist now or may come to our systems in future. (3) real time. This module spends less price, response quickly, has the least impaction to our programs. (4) modulization. Insmod and rmmod is all we need to do, no other configuration involved.This program appears as kernel module. After got the sys_call_table we can replace the corresponding system calls with our ones according to the call number. Then I record the useful information first and call original function next to complete this call. My program reserve two structs, one is used to record socket information. Another is used to record network throughput per process. Implement throughput control by making the process which excess the max passable count slept until a certain time elapse. Test shows this method has reached the system requirements, can control the network throughput effectively according to set policy.