Thiebaut D., Stone H. Improving Disk Cache Hit-Ratios Through Cache Partitioning IEEE Transactions on Computers, vol. 41, No.6, June 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the cache allocations when multiple or non feasible cache allocations exist. This criteria is used by an offline cache allocation algorithm as follows. 2. 4 Offline Cache Allocation Offline procedures for static cache partitioning have been considered in the past, for example, in [10] and [13, 15]. These papers, however, did not consider QoS and were interested primarily in maximizing the hit ratio. The fundamental idea in [13, 15] is to look at the hit ratio versus cache size curves for all the classes where LRU is used as the replacement policy. They seek to find a cache size for each ....

....algorithm as follows. 2.4 Offline Cache Allocation Offline procedures for static cache partitioning have been considered in the past, for example, in [10] and [13, 15] These papers, however, did not consider QoS and were interested primarily in maximizing the hit ratio. The fundamental idea in [13, 15] is to look at the hit ratio versus cache size curves for all the classes where LRU is used as the replacement policy. They seek to find a cache size for each curve such that the hit ratio derivatives for each curve at its chosen cache size are equal. Essentially, the same idea is considered in ....

[Article contains additional citation context not shown here]

D. Thiebaut, H. S. Stone, and J. L. Wolf. Improving disk cache hit-ratios through cache partitioning. IEEE Tran. Comp., 41(6), June 1992.

....absolute and marginal. An analysis of absolute CBA was also presented there. Our research follows [15] 18] and [7] we present an algorithm for file prefetching using marginal cost benefit analysis. This approach employs the concept of marginal gains, similar to the approach used in [14] [17] and [8] Marginal gains are the benefits obtained by taking actions earlier than they otherwise would have been taken, or allocating more resources than otherwise would have been allocated. Previous researchers partitioned the file cache and used marginal gains to establish how many buffers to ....

....the benefits obtained by taking actions earlier than they otherwise would have been taken, or allocating more resources than otherwise would have been allocated. Previous researchers partitioned the file cache and used marginal gains to establish how many buffers to assign to each partition. In [17] each partition corresponded to a different process, while in [8] each partition corresponded to a different access pattern. Our approach uses the concept of marginal gains on a per block basis so that the most time critical blocks are prefetched. Prefetched Balance Figure 1. CBA: seeking a ....

Thiebaut, Dominique, Harold S. Stone and Joel L. Wolf. Improving Disk Cache Hit-Ratios Through Cache Partitioning. IEEE Transactions on Computers, Vol. 41, No. 6. June, 1992

....area. However, the partitioning was performed by collecting the miss rate information of each process o line, and we did not describe techniques to partition the cache memory at run time. Thi ebaut, Stone and Wolf applied their theoretical partitioning study [13] to improve disk cache hit ratios [18]. The model for tightly interleaved streams is extended to be applicable for more than two processes. They also describe the problems in applying the model in practice, such as approximating the miss rate derivative, non monotonic missrate derivatives, and updating the partition. Trace driven ....

Thiebaut, D., H. S. Stone, and J. L. Wolf: 1992, `Improving Disk Cache HitRatios Through Cache Partitioning'. IEEE Transactions on Computers 41(6).

....their policy simply assumes the probability for a evicted block to be accessed in the next time quantum as a constant, which is neither validated nor is it described how this probability is obtained. Thiebaut, Stone and Wolf applied their partitioning work [15] to improve disk cache hit ratios [21]. The model for tightly interleaved streams is extended to be applicable for more than two processes. They also describe the problems in applying the model in practice, such as approximating the miss rate derivative, non monotonic miss rate derivatives, and updating the partition. Trace driven ....

D. Thiebaut, H. S. Stone, and J. L. Wolf. Improving disk cache hit-ratios through cache partitioning. IEEE Transactions on Computers, 41(6), June 1992.

....improvement by partitioning. Since each process may not need the entire cache space, the effect of context switches can be mitigated by keeping useful data in the cache over context switches. It is crucial for modern microprocessors to minimize inter process conflicts by proper cache partitioning [21, 9] or scheduling [17, 23] Our model requires information that is relatively easy to acquire. The characteristics for each process are given by the miss rate as a function of cache size when the process is isolated, which can be easily obtained either on line or offline. The time quantum for each ....

....their policy simply assumes the probability for a evicted block to be accessed in the next time quantum as a constant, which is neither validated nor is it described how this probability is obtained. Thi ebaut, Stone and Wolf applied their partitioning work [18] to improve disk cache hit ratios [21]. The model for tightly interleaved streams is extended to be applicable for more than two processes. They also describe the problems in applying the model in practice, such as approximating the miss rate derivative, non monotonic miss rate derivatives, and updating the partition. Trace driven ....

D. Thi'ebaut, H. S. Stone, and J. L. Wolf. Improving disk cache hit-ratios through cache partitioning. IEEE Transactions on Computers, 41(6), June 1992.

....However, the partitioning was performed by collecting the miss rate information of each process off line. The work of [10] did not investigate how to partition the cache memory at run time. Thiebaut, Stone and Wolf applied their theoretical partitioning study [9] to improve disk cache hit ratios [11]. The model for tightly interleaved streams is extended to be applicable for more than two processes. They also describe the problems in applying the model in practice, such as approximating the miss rate derivative, non monotonic missrate derivatives, and updating the partition. Trace driven ....

....for 32 MB disk caches show that the partitioning improves the relative hit ratios in the range of 1 to 2 over the LRU policy. Our partition work differs from previous efforts. It works for set associative caches with multiple threads and a coarse grained partition, whereas Thiebaut et al. [11] only focused on disk caches that are fully associative with cache block granularity. Finally, this work discusses an on line method to partition the cache, whereas our previous only covered partitioning based on off line profiling [10] 3. Partitioning Algorithm This section presents our cache ....

D. Thiebaut, H. S. Stone, and J. L. Wolf. Improving disk cache hit-ratios through cache partitioning. IEEE Transactions on Computers, 41(6), June 1992.

....managed separately. An important problem that should be addressed then is how to allocate the blocks in the cache among the three partitions. 6 To this end, we use the notion of marginal gains, which has frequently been used in resource allocation strategies in various computer systems areas [8, 11, 12, 13]. Marginal gain is defined as MG(n) Hit(n) Gamma Hit(n Gamma 1) which specifies the expected number of extra buffer hits per unit time that would be obtained by increasing the number of allocated buffers from (n 1) to n, where Hit(n) is the expected number of buffer hits per unit time using n ....

....gain is MG loop (n) 1=p i max 1 . Finally, for the partition that holds the other references and which is managed by the LRU replacement scheme, the expected number of buffer hits per unit and the expected marginal gain value can be calculated from the buffer hit ratio using ghost buffers [11, 8] and or the Belady s lifetime function [14] Prediction of buffer hit ratio using only ghost buffers is impractical due to the overhead of measuring the hit counts of all LRU stack positions individually. In the UBM scheme, we use an approximation method suggested by Choi et al. 13] The proposed ....

D. Thiebaut, H. S. Stone, and J. L. Wolf. Improving Disk Cache Hit-Ratios Through Cache Partitioning. IEEE Transactions on Computers, 41(6):665--676, 1992.

....being managed separately. An important problem that should be addressed then is how to allocate the blocks in the cache among the three partitions. To this end, we use the notion of marginal gains, which has frequently been used in resource allocation strategies in various computer systems areas [10, 11, 12, 13]. Marginal gain is defined as MG(n) Hit(n) Hit(n ; 1) which specifies the expected number of extra buffer hits per unit time that would be obtained by increasing the number of allocated buffers from (n 1) to n, where Hit(n) is the expected number of buffer hits per unit time using n buffers. ....

....gain is MG loop (n) 1=p i max 1 . Finally, for the partition that holds the other references and which is managed by the LRU replacement scheme, the expected number of buffer hits per unit time and the expected marginal gain value can be calculated from the buffer hit ratio using ghost buffers [11, 10] and or the Belady s lifetime function [14] Ghost buffers, sometimes called dataless buffers, are used to estimate the number of buffer hits per unit time for cache sizes larger than the current size when the cache is managed by the LRU replacement scheme. A ghost buffer does not contain any data ....

D. Thiebaut, H. S. Stone, and J. L. Wolf. Improving Disk Cache Hit-Ratios Through Cache Partitioning. IEEE Transactions on Computers, 41(6):665--676, 1992.

....(i.e. the time to resolve a name in X.400) the size of the item, etc. Another way is to divide the cache into partitions and to allocate a partition to names belonging only to a specific class. Recent results indicate that cache partitioning increases performance in disk and database systems [6, 13]. In this Section, we describe the results obtained with cache partitioning in X.500. We identified earlier 4 classes of names. Therefore, it is natural to divide the cache into 4 partitions. We partition the cache in a mathematical sense, i.e. we divide it into disjoint areas whose union forms ....

....partitioned and the non partitioned cache are essentially undistinguishable. In fact, it turns out that the miss ratio is slightly higher for the partitioned cache. We observed a similar result with the trace collected at INRIA Sophia [3] This is also in agreement with other results reported in [6, 13]. 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 16.00 18.00 0.00 10.00 20.00 30.00 40.00 50.00 3 Misses x 10 Messages x 10 3 Partitioned cache Non partitioned cache Figure 6: Miss histories for the partitioned and the non partitioned cache Figure 7 shows the weighted miss history ....

[Article contains additional citation context not shown here]

Thiebaut D., Stone H. "Improving disk cache hit ratios through cache partitioning", IEEE Transactions on Computers, Vol. 41, No. 6, pp. 665-676, June 1992.

....name resolution with the partitioned cache can be twice as low that obtained with the non partitioned cache. The fact that cache partitioning does increase performance is expected since similar results have been observed in both disk and file systems using analytic and simulation approaches (e.g. [27]) However, the huge performance increase which is obtained in our case is somewhat surprising. We are currently investigating this issue. 4 Conclusion In this paper, we have evaluated the impact of name caching on the performance of Pizarro, our implementation of the OSI Directory. We have ....

D. Thiebaut, H. S. Stone, J. L. Wolf, "Improving disk cache hit-ratios through cache partitioning", IEEE Trans. Computers, vol. 41, no. 6, pp. 665-676, June 1992.

No context found.

Thiebaut D., Stone H. Improving Disk Cache Hit-Ratios Through Cache Partitioning IEEE Transactions on Computers, vol. 41, No.6, June 1992.

No context found.

D. Thiebaut, H.S. Stone, and J.L. Wolf. Improving disk cache hit-ratios through cache partitioning. IEEE Transactions on Computers, 41:665--676, 1992.

No context found.

D. Thiebaut, H.S. Stone, and J.L. Wolf. Improving disk cache hit-ratios through cache partitioning. IEEE Transactions on Computers, 41:665--676, 1992.

No context found.

D. Thiebaut, H. S. Stone, and J. L. Wolf. Improving disk cache hit-ratios through cache partitioning. IEEE Trans. Comput., 41(6):665--676, 1992.

No context found.

D. Thiebaut, H. Stone, J. Wolf. Improving Disk Cache Hit-Ratios Through Cache Partitioning, IEEE Transactions on Computers, Vol. 41, No. 6, June 1992.

No context found.

Dominique Thiebaut, " Improving Disk cache Hit Ratio Through Cache Partitioning" , IEEE Transaction on Computers, Vol.41,No.6, June1992.

No context found.

D. Thibaut and H. S. Stone. Improving disk cache hit-ratios through cache partitioning. IEEE Transactions on Computers, 41(6):665--676, 1992.

No context found.

Thiebaut D., Stone H. Improving Disk Cache Hit-Ratios Through Cache Partitioning IEEE Transactions on Computers, vol. 41, No.6, June 1992.

No context found.

Thiebaut, D., Stone, H., and Wolf, J. Improving Disk Cache Hit-Ratios Through Cache Partitioning. IEEE T'atsactiots o Computers, Vol. 41, No. 6, pp. 665 - 676, June, 1992.

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