Assignment1 - CSG1102: Virtual Memory. Christoer V. Hallstensen snr: March 28, 2011

Transcription

1 Assignment1 - CSG1102: Virtual Memory Christoer V. Hallstensen snr: March 28,

2 Contents 1 Abstract 3 2 Virtual Memory with Pages Virtual memory management Virtual memory support for multiple page sizes Page replacement in distributed virtual memory systems Performance measurement and modeling to evaluate various effects on a shared memory multiprocessor Static grouping of small objects to enhance performance of a paged virtual memory Virtual Memory with Segmentation Segmentation A fast translation method for paging on top of segmentation Virtual Memory Virtual memory, processes, and sharing in MULTICS Generic virtual memory management for operating system kernels 7 References 8 2

3 1 Abstract This is an assignment for school in the unit CSG Operating Systems about making a personal reference library about Virtual Memory. It contains 10 references of books, journal papers and conference papers with a short one paragraph summary of most of them. The document is split up in two parts; Virtual Memory with Paging and Virtual Memory with Segmentation. 3

4 2 Virtual Memory with Pages 2.1 Virtual memory management (Deitel, Deitel, & Chones, 2004) A page replacement strategy is a selected method on how a computer system determine which page to swap when the main memory is full. There are several page replacement strategies that can be applied like Random, First-In- First-Out, Least-Recently-Used, Least-Frequently-Used and Not-Used-Recently. Each of them has advantages and disadvantages like for instance; Random is fast, but it does not care if the page replaced is used frequently, so it can replace a page often accessed and make unnecessary overhead. FIFO swaps the rst page regardless of it is the most used page in memory. LRU keeps the pages that are often used but use some CPU time calculating the LRU page, this can take time in large memory systems. When the memory manager tries to grab a page that does not exist in virtual memory a page-fault occurs. This puts the process requesting the page on I/O wait until the correct page can be retrieved from secondary storage. To speed up things modern architecture uses a modied bit so the operating system easily can determine if the page has been updated or modied recently. 2.2 Virtual memory support for multiple page sizes (Khalidi, Talluri, Nelson, & Williams, 1993) This conference paper is about how to make use of future hardware more ecient by using multiple page sizes instead of one xed page size. It discusses potential advantages by redesigning virtual memory in operating systems to support the hardware implementation of multiple page sizes and addresses some potential key issues of implementation regarding page addressing, allocation and replacement. 4

5 2.3 Page replacement in distributed virtual memory systems (Malkawi, Knox, & Abaza, 1992) This conference paper from IEEE looks at page replacement designs and implementation in larger distributed computer systems like supercomputers. Distributed systems are much more complex since the page requested can be located on other physical machines in the system. Distributed Systems operate with three types of virtual memory pages; local, global and remote pages compared to conventional computers that only has local page tables. For instance Process A running on Node A can own a page table located at Node B. This makes conventional page placement/replacement harder. While Least Recently Used page can easily be determined on the local system, it is hard to do on globally and remote memory. Instead of using LRU, they use Least Recently Brought on global and remote memory. This algorithm replaces the page that has been stored on the node the longest with the new one. 2.4 Performance measurement and modeling to evaluate various eects on a shared memory multiprocessor (Zhang, 1991) Brings up the need for on-demand paging in a multiprocessor environment. 2.5 Static grouping of small objects to enhance performance of a paged virtual memory (Stamos, 1984) 5

6 3 Virtual Memory with Segmentation 3.1 Segmentation (Tanenbaum, 2009) Segmentation is practical when a program require large dynamic memory allocation. For instance when a program is to large to t in one large chunk, it can be segmented into pieces to use the memory more ecient by putting blocks of code in free available spaces. These segments can vary in size, and may change size while executing. The segments live in their own address space which give them the room to expand if needed without interfering with each other. A segment can consist of dierent data types or a procedure but this is not normal. Because segments live in their own space in memory with a unique segment address, linking procedures or sharing data or procedures between multiple processes are much more simplied. An example of this is the use of DLL's or libraries, where code is used in several processes at once. This also means that if one module is changed in a program, only that module have to be recompiled and some memory addresses may need to be changed, not the whole program. The real dierence between pages and segments is that pages are static while segments are dynamic. 3.2 A fast translation method for paging on top of segmentation (Dally, 1992) This paper propose a faster method of translating page addresses within a segment that only need one read compared to the conventional method that requires several translation to locate a page within a segment. In the traditional way of doing it the rst step is to check segment bounds and access privileges. If validated the next step is to get a page translation. The faster method is looking up both the segment and page in a combined segment/page descriptor. 3.3 Virtual Memory (Denning, 1970) Operating systems today strive to give programmers four objectives; modularity, ability to share resources, varied size of data structures and program security. These four elements can be achieved by using segmentation of virtual memory. By modularizing the program in segments, each data structure, procedure and scalars can have their own re-sizable address space. This gives the programmer the the three rst objectives. Program security can be achieved by adding a protection bit in the memory allocation table by setting permissions for read, write or fetching instructions from the selected memory space. 6

7 3.4 Virtual memory, processes, and sharing in MULTICS (Daley & Dennis, 1968) This paper explains the design features of MULTICS that make it able to serve several users running diverse applications in their own address space. This is done in the MULTICS environment by using segmentation. On a MULTICS system each programmer has his own address space or segment and a process runs in one-to-one mapping with virtual memory. Because of the need to share procedures, each procedure is mapped in its own segment for easy access by other processes or programmers. Each segment is distinct from each other and live in their own space with their own attributes and access control. They have the ability to increase or decrease in size without depending on anything outside. 3.5 Generic virtual memory management for operating system kernels (Abrossimov, Rozier, & Shapiro, 1989) This paper talks about memory abstractions by using a general memory interface with a practical approach on how to use virtual memory. 7

8 References Abrossimov, E., Rozier, M., & Shapiro, M. (1989). Generic virtual memory management for operating system kernels. In Proceedings of the twelfth acm symposium on operating systems principles (pp ). New York, NY, USA: ACM. Daley, R. C., & Dennis, J. B. (1968, May). Virtual memory, processes, and sharing in multics. Commun. ACM, 11, Dally, W. (1992, February). A fast translation method for paging on top of segmentation. Computers, IEEE Transactions on, 41 (2), Deitel, H., Deitel, P., & Chones, D. (2004). Operating systems (3 ed.). Pearson Education. Denning, P. J. (1970, September). Virtual memory. ACM Comput. Surv., 2, Khalidi, Y., Talluri, M., Nelson, M., & Williams, D. (1993, October). Virtual memory support for multiple page sizes. In Workstation operating systems, proceedings., fourth workshop on (p ). Malkawi, M., Knox, D., & Abaza, M. (1992, December). Page replacement in distributed virtual memory systems. In Parallel and distributed processing, proceedings of the fourth ieee symposium on (p ). Stamos, J. W. (1984, May). Static grouping of small objects to enhance performance of a paged virtual memory. ACM Trans. Comput. Syst., 2, Tanenbaum, A. S. (2009). Modern operating systems (3 ed.). Pearson Education. Zhang, X. (1991, January). Performance measurement and modeling to evaluate various eects on a shared memory multiprocessor. Software Engineering, IEEE Transactions on, 17 (1),