Alan Jay Smith. Disk cache-miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161-203, August 1985.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....cycles removed. In this chapter we have presented an overview of MPEG applications. We have broken down the applications both by software functional block and by architectural component. In the next chapter, we investigate I O in greater detail. IO Disk caches have been thoroughly reported in [Smi85] and Ousterhout [ODH 85] A great body of work exists on file systems for multimedia systems. Examples include [RW93] RV91] and [AOG92] In general these works assume a complicated client server I O model with multiple disks, multiple clients, and several simultaneous I O sessions. Issues ....

Alan Jay Smith. Disk cache--miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....algorithms for a simple graph: sequences of pages. Part of our future work is to investigate applying the general access graph algorithms for more complicated patterns in applications. Along with the theoretical studies there have been extensive experimental studies. A good survey can be found in [12, 32]. Except a few early studies [22] most experimental studies focused on efficient implementation of LRU page replacement algorithms in practice. A number of approximation algorithms, FIFO with second chance [4] CLOCK algorithms [3] etc. have been proposed and implemented in various systems [8, ....

Alan J. Smith. Disk cache---miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....I O traces or installing instrumentation. Traces are needed by all designers to simulate and directly compare various cache alternatives and understand workload locality. Generating traces is difficult, and the type of tracing or instrumentation determines which I O characteristics are visible [15]. Some I O cache studies have used disk requests to study cache performance [15, 10, 11] while others have used operating system traces of file system activity [9, 8, 3, 13, 6, 12] Figure 1 describes the eleven workloads evaluated in this paper. Most of the workloads cover several days of user ....

....to simulate and directly compare various cache alternatives and understand workload locality. Generating traces is difficult, and the type of tracing or instrumentation determines which I O characteristics are visible [15] Some I O cache studies have used disk requests to study cache performance [15, 10, 11], while others have used operating system traces of file system activity [9, 8, 3, 13, 6, 12] Figure 1 describes the eleven workloads evaluated in this paper. Most of the workloads cover several days of user activity. These long traces capture significant I O activity and show the interaction of ....

Alan Jay Smith. Disk cache - miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....hold copies of recently referenced words from main memory. So far, we have discussed caching only in the context of managing the memory space of a processor. Many other forms of caching exist. Caches of recently referenced disk blocks held in main memory increase overall disk system performance [58, 51]. Digital typesetters cache font information to reduce the amount of data transmitted over the communications channel from the host computer [23] Program execution times can be enhanced by precomputing and caching values of expensive functions (e.g. trigonometric functions) and using table ....

....had been accessed [53] Satyanarayanan analyzed file sizes and lifetimes on a PDP 10 system [46] but the study was made statically by scanning the contents of disk storage at a fixed point in time. More recently, Smith used trace data from IBM mainframes to predict the performance of disk caches [51]. Four recent studies contain UNIX measurements that partially overlap ours: Lazowska et al. analyzed block size tradeoffs in remote file systems, and reported on the disk I O required per user [32] McKusick et al. reported on the effectiveness of current UNIX disk caches [36] and Ousterhout et ....

[Article contains additional citation context not shown here]

Alan Jay Smith. Disk Cache - Miss Ratio Analysis and Design Considerations. ACM Transactions on Computer Systems 3(3):161-203, August, 1985.

....5 Related Work Memory hierarchies and virtual memory systems have been studied in great detail. Earlier work focused on uniprocessor architectures, with emphasis on multilevel memory hierarchies [2, 25, 32, 23] memory access patterns [1] virtual memory systems [10, 11] memory and disk caches [33, 34]. Most related work on memory management for parallel architectures includes page placement strategies for distributed shared memory architectures, and shared virtual memory systems. Non uniform memory access (NUMA) architectures allow processors to access data in either local or remote memory ....

Alan J. Smith. Disk Cache---Miss Ratio Analysis and Design Considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....virtual memory systems provide software management of memory accesses, it is natural to use these facilities to ensure that accesses to shared data are performed consistently. We note that many existing parallel programming paradigms use virtual memory pages for interprocessor communication [Smi85b, LH86, RT 88, CF89, LH89, LE90] In this section, we first show that selective write through is an effective approach to the stale dual port memory data problem. Then, we present two solutions to the stale cache data problem. Selective Write Through Write through caches have often been ....

Alan Jay Smith. Disk cache---miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....cache line size. There are a number of cache design parameters that are set either at the time that the device is manufactured or are set by the system administrator. We discuss these cache parameters in the following sections. Additional information about caches in the I O path can be found in [Brandwajn81, Carson92, Hospodor92, Hospodor94, Karedla94, Smith85]. In particular, Brandwajn81] gives a good description of the events following cache misses and hits. 23 Multi device HBA Controller Disk Controller Disk Controller Disk Cache Host File system cache controller Figure 2.4: Possible locations in the I O path for the cache. Cache ....

....a number of the possible locations for a cache. The cache in the embedded device controller is also used for speed matching between the transfer rate of the disk and the I O channel. Historically, this cache was developed for circumstances where rotational position sensing (RPS) misses occur [Hou92, Smith85]. In practice. Almost all current disk drives have caches in the embedded controllers [Karedla94] Disk arrays invariably have caches in the array con24 troller. Operating system cache support is pervasive. Caches in the HBA tend to be optional. Cache hardware configuration The cache hardware ....

[Article contains additional citation context not shown here]

Alan Jay Smith. Disk cache---miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....for each component. Different components require different block and cache sizes to capture locality effectively. Understanding the sources of locality and the cache properties that best capture locality can help to improve I O cache efficiency. For a more thorough study of I O cache behavior see [10, 9]. If all I O requests are cached without regard to file type, few options exist for reducing the cache misses and improving cache performance. Figure 8 shows the cache miss behavior of the Application workload in a unified cache that uses no information about request types. The miss ratio is the ....

Alan Jay Smith. Disk cache - miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....which may consume immense amounts of data. Disk I O has always been slower than processing speed, and recent trends have shown that improvements in the speed of disk hardware are not keeping up with the increasing raw speed of processors. This widening access time gap is known as the I O crisis [14, 20]. The problem is compounded in typical parallel architectures that multiply the processing and memory capacity without balancing the I O capabilities. The most promising solution to the I O crisis is to extend parallelism into the I O subsystem. One such approach is to connect many disks to the ....

....from disk striping. Examples of this I O architecture include the Concurrent File System [15, 6] for the Intel iPSC 2 multiprocessor, and the Bridge file system [4, 3] for the BBN Butterfly multiprocessor. Caching commonly used disk blocks can significantly improve file system performance [20], and indeed is a technique used in most modern file systems. Prefetching is also successful in uniprocessor file systems [20, 18, 19, 17] The central idea behind prefetching is to overlap some of the I O time with computation by issuing disk operations before they are requested. With parallel ....

[Article contains additional citation context not shown here]

Alan Jay Smith. Disk cache-miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....environment[MDK96] how this will be applicable to buffer cache management in general is still an open question. A different approach which does not require any information neither from the application nor from observations of past behavior is the LRU OBL (One Block Lookahead) policy[Smi78, Smi85] This policy simply prefetches the logical next block of the currently referenced block if it is not resident in cache. It has been shown that through this simple policy, improvements of up to 80 in the hit rate was possible for some workloads[Smi85] The policy that we propose in this paper, ....

....LRU OBL (One Block Lookahead) policy[Smi78, Smi85] This policy simply prefetches the logical next block of the currently referenced block if it is not resident in cache. It has been shown that through this simple policy, improvements of up to 80 in the hit rate was possible for some workloads[Smi85] The policy that we propose in this paper, referred to as the W 2 R (Weighing Waiting Room) policy, starts from the LRU OBL policy and borrows from the 2Q and SLRU policies. It prefetches blocks exactly like the LRU OBL policy requiring minimal overhead in determining the block to prefetch and ....

Alan Jay Smith. Disk cache-miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....which may consume immense amounts of data. Disk I O has always been slower than processing speed, and recent trends have shown that improvements in the speed of disk hardware are not keeping up with the increasing raw speed of processors. This widening access time gap is known as the I O crisis [13, 16]. The problem is compounded in typical parallel architectures that multiply the processing and memory capacity without balancing the I O capabilities. The most promising solution to the I O crisis is to extend parallelism into the I O subsystem. One such approach is to connect many disks to the ....

....on multiprocessor file systems intended for scientific applications. These applications typically push the leading edge of computing technology, such as multiprocessors, placing tremendous demands on both CPU and I O systems. Most file caching studies have examined generalpurpose workloads (e.g. [16]) where files are much smaller [12, 5] The parallel environment and workload raise a number of questions: Are caches useful for parallel scientific applications using parallel file systems If so, in what way What are the appropriate management policies Different workload characteristics, ....

[Article contains additional citation context not shown here]

Alan Jay Smith. Disk cache-miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161--203, August 1985.

....every read and write) generates huge quantities of data, most such studies have produced only summary statistics or made some other compromise such as coalescing multiple I Os together (e.g. Ousterhout85, Floyd86, Floyd89] Many were taken on non UNIX systems. For example: IBM mainframes [Procar82, Smith85, Kure88, Staelin88, Staelin90, Staelin91, Bozman91]; Cray supercomputers [Miller91] Sprite (with no timing data on individual I Os) Baker91] DEC VMS [Ramakrishnan92] TOPS 10 (static analysis only) Satyanarayanan81] The UNIX buffer cache means that most accesses never reach the disk, so these studies are not very good models of what happens ....

Alan Jay Smith. Disk cache---miss ratio analysis and design considerations. ACM Transactions on Computer Systems 3(3):161--203, August 1985.

No context found.

Alan Jay Smith. Disk cache-miss ratio analysis and design considerations. ACM Transactions on Computer Systems, 3(3):161-203, August 1985.

No context found.

Alan Jay Smith. Disk cache---miss ratio analysis and design considerations. ACM Transactions on Computer Systems 3(3):161--203, August 1985.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC