A_survey_paper_on_context_switching.pdf

A Survey Paper on Context Switching Muhammad Q Shahzad CIS-657 Term Paper Syracuse University [email protected]

1.Abstract This paper will highlight the issues pertaining to context switching and some of the ways that we can minimize the impact and create an optimized process execution while doing a context switch that will increase the efficiency based on the previous research. In Hope of doing some of the previous research and conclusions are quoted and don’t give a bright picture in this regard.

2.Introduction: The context switching plays an important part in the process lifecycle and where the processes need to be run and complete quickly and efficiently. Real time environment processes are more time constraint. Real-time systems require the deadline to be achieved for the processes. If the overhead cost is decreased the process will have high performance. Systems failing to meet the constraint can become fatal failure. The context switch is a switch that is required for the queue to keep moving and without this functionality defined, the system could be lost, hence context switch plays an important role in process execution.

3. Context Switch: Switching the processor from the Kernel to the process A and vice versa from process A to the kernel is context switching [1]. In the Operating system it is the switching of the CPU from a process to another process. In embedded systems it is the process of storing the state of the process that can be restored or executed from the same point later, this will allow the processes to share a single CPU. 1

It Saves the registers of the old process in the process table entry for it. Registers R1-R5, the stack pointer, the program counter, and the process status are all saved.. In Xinu Operating system it is identified as ctxsw. ctxsw (oldRegArea, newRegArea),

Process resumption and suspension are implemented by the procedures resume and suspend. it saves the return address of ctxsw. Thus, when the old process is resumed, it starts executing after the statement in Resched that calls ctxsw. The stack pointer must be adjusted to make it look as if a return from ctxsw occurred. The diagram below illustrates the process of context switching.

Figure 1: Example of a context switch

2

3. Context Switch Process improvement analysis: There are multiple analysis that have been done to show the possibility of the improvement in performance in context switching and the ways it affects the process and the processor, the time it will take , the clock cycle , input and output from the user, all of these will be taken into effect. The processes that are less CPU intensive and are also called lightweight processes will share the address space with the high CPU intensive processes . The context switch between two lightweight processes will reduce in this environment will reduce the amount of state necessary to save and restore [2] but it will still cost to save and restore. According to another research done by Agarwal[3] large caches suffered considerable performance loss when forced to purge their contents after each context switch, whereas small caches did not suffer as much. As in the research above with the lightweight and the heavy weight processes, it correlates this research.

There is another possibility that a data flow analysis that will could be done to identify the dead registers that are not in use will introduce a concept of Fast context switching that will in turn provide more efficiency with register remapping that is used by a live range optimization allocated to a scratch register. The register transfer represents a change in the machine's state, and instructions which cause more than one state change are represented by lists of register transfers. [4] In this research [6],direct time cost per context switch where processes make no data access was measured and the the total time cost per context switch that includes the direct cost of restoring cache state. The overall data size has a substantial impact on the cost of the context if it is larger then the cache size [5], based on this study , the environment where this research was done played a significant role as well.As illustrated in the final execution graph below that identifies the performance of the context switch when the Array size is in between 0 -2048 with cache size of 512 KB and the context switch time ranges from 0 - 1500

3

The effect of access stride on the cost of context switch

To increase the efficiency for the [7]context switching selective context switching is utilized as it makes decision at the point between a request for handling I/O requests in the process context and the activation of the device in the kernel context. An illustration of this concept can be seen below:

Figure 3 : Selective Context Switching

ARM Cortex-M3 processor achieves a shorter context switch time [8]. In this research The figure below illustrates the improvement in time by using this processor and also g Task Yielding in a Priority-based Scheduler.This processor uses a hardware based trap for the context switching , the software traps have their own overheads. But using the task yielding algorithm with the hardware trap the performance was increased.

4

Figure 4: Experimental result of context switching

This task yielding algorithm that was utilized in the study is as follow:

Figure 5: Task yielding algorithm

Conclusion: Based on the papers surveyed above and the research the context switching is expensive and the cost is really difficult to reduce, the CPU overhead cannot be reduced but using multiple threads can decrease only the saving and restoring the register states but will still impact the overall performance. In the selective context switching scheme that increase the performance , but flash memory was utilized. Also when the hardware based trap for context based switching is utilized the performance is greatly increased. If the performance of the context switching can be 5

increased based on the Operating systems , it will result in the system performance that we have never imagined

Reference: [1] An Operating Systems Vade Mecum Raphael A. Finkel University of Wisconsin at Madison [2] Shem-Tov Levi and Ashok K. Agrawala, Real-Time Systems Design, McGraw{Hill, 1990. [3] Gary J. Nutt, Centralized and Distributed Operating Systems, Prentice{Hall, Inc., 1992. [4] Anant Agarwal, John Hennessy, and Mark Horowitz, \Cache Performance of Operating System and Multiprogramming Workloads", ACM Transactions on Computer Systems, November 1988, 6(4):393{431. [5] Jack W. Davidson and David B. Whalley, \A Design Environment for Addressing Architecture and Compiler Interactions", Microprocessors and Microsystems, November 1991, 15(9):459{472. [6] Chuanpeng Li,Chen Ding,Kai Shen \ Quantifying The Cost of Context Switch ,” D.4.8 [Operating Systems]: Performance—Measurements; C.4 [Computer Systems Organization]: Performance of Systems—Measurement techniques” [7] Sehwan Lee, Seunghwan Hyun, \ Selective Context Switching Scheme on Flash Memory , “2009 International Conference on Computational Science and Its Applications” [8] Evaluations of Hardware and Software-Based Context Switching Methods in Cortex-M3 for Embedded Applications \ Hayeon Choi and Sangsoo Park , ” International Journal of Smart Home Vol. 9, No. 2 (2015), pp. 111-122 “

6