I. INTRODUCTIONFailures are prevalent in large-scale storage clusters 的简体中文翻译

I. INTRODUCTIONFailures are prevale

I. INTRODUCTIONFailures are prevalent in large-scale storage clusters andmanifest at disks or various storage components [10], [15], [30],[36]. Field studies report that disk replacements in productionare more frequent than estimated by vendors [30], [36], andlatent sector errors are commonly found in modern disks [4].To maintain data availability guarantees in the face of failures,practical storage clusters often stripe data with redundancyacross multiple nodes via either replication or erasure coding.Replication creates identical data copies and is adopted byearlier generations of storage clusters, yet it incurs substantialstorage overhead, especially with today’s tremendous growthof data storage. On the other hand, erasure coding creates alimited amount of redundant data through coding computations,and provably maintains the same level of fault tolerance withmuch less storage redundancy than replication [41]. Today’slarge-scale storage clusters increasingly adopt erasure codingto provide low-cost fault-tolerant storage (e.g., [2], [10], [13],[26], [28]), and reportedly save petabytes of storage comparedto replication [13], [26].While being storage-efficient, erasure coding incurs highrepair penalty. As an example, we consider Reed-Solomon (RS)codes [34], which are a popular erasure coding constructionused in production [2], [10], [26], [28]. At a high level, RScodes encode k data chunks into n coded chunks for someparameters k and n > k, such that any k out of n codedchunks can reconstruct (or decode) all original k data chunks.However, repairing a lost chunk of RS codes needs to retrievek available chunks for decoding, implying that both bandwidthand I/O costs for a single-chunk repair are amplified k times;in contrast, in replication, repairing a lost chunk can be simplydone by retrieving another available chunk copy.The high repair penalty is a fundamental issue in all erasurecoding constructions: the repair traffic increases as the storageredundancy decreases [8]. Thus, there have been extensivestudies on improving the repair performance of erasure coding,such as proposing theoretically proven erasure codes thatminimize the repair traffic or I/Os during repair (e.g., [8],[13], [35]), or designing repair-efficient techniques that applyto all practical erasure codes including RS codes (e.g., [5],[20], [21], [24], [37], [38]). Conventional repair approachesare reactive, meaning that a repair operation is triggered onlyafter a node failure is detected. Nevertheless, if we can predictimpending failures in advance, we may proactively repair thelost data of any impending failed node to mitigate the repairpenalty before any actual failure occurs.Recent studies show that machine learning can achieveaccurate prediction of disk failures in production environmentswith thousands of disks [6], [18], [23], [42], [43], [45]; in somecases, the prediction accuracy can even reach at least 95% [6],[18], [23], [45] with a very small false alarm rate. Motivated bythe potential of highly accurate disk failure prediction, we canaccurately pinpoint a soon-to-fail (STF) node and accelerate arepair operation by coupling two repair methods: (i) migration,in which we relocate the currently stored chunks of the STFnode to other healthy nodes, and (ii) reconstruction, in whichwe reconstruct (or decode) the chunks of the STF node byretrieving the chunks of all healthy nodes in a storage cluster asin conventional reactive repair approaches. Migration addressesthe bandwidth and I/O amplification issues that are inherentin erasure coding, while reconstruction exploits the aggregatebandwidth resources of all healthy nodes. An open question ishow to carefully couple both migration and reconstruction soas to maximize the repair performance.We present FastPR, a Fast Predictive Repair approach thatcarefully couples the migration and reconstruction of the chunksof the STF node, with the primary objective of minimizingthe total repair time. FastPR schedules both migration andreconstruction of the chunks of the STF node in a parallelfashion, so as to exploit the available bandwidth resources ofthe underlying storage cluster. We address two repair scenarios:scattered repair, which stores the repaired chunks of the STFnode across all other existing nodes in the storage cluster,and hot-standby repair, which stores the repaired chunks ofthe STF node in dedicated hot-standby nodes. We present anin-depth study of FastPR through mathematical analysis, largescale simulation, and Amazon EC2 experiments, and make thefollowing contributions:• We first present mathematical analysis on the optimalpredictive repair in minimizing the total repair time. We show
0/5000
源语言: -
目标语言: -
结果 (简体中文) 1: [复制]
复制成功!
I.引言<br>故障是在大规模存储集群流行和<br>清单在磁盘或各种存储组件[10],[15],[30],<br>[36]。现场研究报告在生产磁盘更换<br>更加频繁比厂商[30],[36]估计,和<br>潜在扇区错误在现代磁盘[4]常见。<br>在故障,面对保持数据的可用性保证<br>实际存储簇往往具有冗余磁条数据<br>经由任一复制或擦除编码跨多个节点。<br>复制将创建相同的数据拷贝,并通过采用<br>存储集群的前几代,但它会带来可观<br>的开销存储,尤其是今天的巨大增长<br>的数据存储。在另一方面,擦除编码创建一个<br>经过编码的计算的冗余数据的数量有限,<br>并且可证明保持容错与同级别<br>少得多的存储冗余比复制[41]。今天的<br>大型存储集群越来越多地采用擦除编码<br>,以提供低成本的容错存储(例如,,,[13],[2] [10] <br>,[26] [28]),并且据报道节省存储的PB的比较<br>到复制[13],[26]。<br>虽然是高存储效率,擦除编码导致很高的<br>维修处罚。作为一个例子,我们考虑的Reed-Solomon(RS)<br>码[34],这是一个受欢迎的擦除编码结构<br>在生产[2]所使用的,[10],[26],[28]。在高层次上,RS<br>码编码k个数据块成一些n个编码块<br>参数k和N> K,使得任何K掉的n个编码<br>块可以重构(或解码)所有原始k个数据块。<br>然而,修复的RS码需要一个丢失块检索<br>ķ可用块用于解码,这意味着这两个带宽<br>为单块修复和I / O开销被放大k倍; <br>相反,在复制,修复丢失的块可以简单地<br>通过检索另一个可用的块复制完成。<br>高修处罚是所有擦除的一个基本问题<br>编码结构:修流量的增加作为存储<br>冗余降低[8]。因此,已经有大量的<br>改善擦除的编码,修复性能的研究<br>如提议理论上证明纠删码<br>最小化修复在修复流量或I / O的(例如,[8],<br>[13],[35]),或设计修复效技术适用<br>于所有实际擦除码包括RS码(例如,[5],<br>[20],[21],[24],[37],[38])。常规修复方法<br>是反应性的,这意味着只有修复操作被触发<br>检测节点故障后。不过,如果我们可以预测<br>提前即将发生的故障,我们会主动修复<br>任何即将发生的故障节点的丢失的数据,以减轻修复<br>发生任何实际故障之前的处罚。<br>最近的研究表明,机器学习可以实现<br>的磁盘故障准确预测在生产环境中<br>与成千上万盘[6],[18],[23],[42],[43],[45]; 在一些<br>情况下,预测精度甚至可以达到至少95%[6],<br>具有非常小的误报率[18],[23],[45]。通过激励<br>高精度的磁盘故障预测的潜力,我们可以<br>精确地查明一个即将失效(STF)节点和加速<br>通过耦合两个修理方法修理操作:(ⅰ)迁移,<br>在其中我们重新定位的当前存储的数据块在STF <br>节点到其他节点的健康,以及(ii)重建,其中<br>我们重构(或解码)STF节点的由块<br>检索存储在存储簇中的所有健康的节点的块作为<br>在常规反应性修复的方法。迁移地址<br>是固有的带宽和I / O放大问题<br>在擦除编码,而重建利用聚合<br>所有健康节点的带宽资源。一个悬而未决的问题是<br>如何认真夫妇的迁移和重建,从而<br>为最大限度地修复性能。<br>我们目前FastPR,快速修复预测做法,<br>认真夫妇迁移和块的重建<br>的STF节点,其首要的目的减少的<br>总维修时间。FastPR时间表迁移和<br>在STF节点的组块的重建中的并联<br>方式,以便利用的可用带宽资源<br>的基础存储群集。我们解决两个维修方案:<br>散修,哪些商店的STF的修复块<br>的存储集群中的所有其他现有节点的节点,<br>和热备份修复,哪些商店的修复块<br>专用热备用节点的STF节点。我们提出了一个<br>FastPR的深入研究,通过数学分析,大规模仿真,以及亚马逊EC2的实验,并进行?<br>以下贡献:<br>•我们在最佳的第一本数学分析<br>预测维修最小化总维修时间。我们展示
正在翻译中..
结果 (简体中文) 2:[复制]
复制成功!
Health and behavioral Education The Karen tribes found that the tribes Karen had The culture of food consumption differs from the northern indigenous and other surgeries in Doi Luang district. There are also other cultural beliefs that are linked to self-care, helping to promote Prevention of diabetes (health belief model) is said to be a health belief based on the practice of individuals in the prevention of disease or cooperation in the prevention of the disease, and the contributing factors that lead to the need for the health of the community. From the above situation, The researchers studied health and behavior prevention of diabetes in the ethnic groups of Doi Luang. The province of Xiangtan to bring out the results Guidelines for the design of health care systems to prevent the occurrence of diabetes ...
正在翻译中..
结果 (简体中文) 3:[复制]
复制成功!
一、 导言<br>故障在大型存储群集和<br>在磁盘或各种存储组件上的清单[10]、[15]、[30],<br>[36]。现场研究报告,生产中的磁盘更换<br>比供应商估计的更频繁[30],[36],以及<br>潜在的扇区错误通常出现在现代磁盘中[4]。<br>为了在出现故障时保证数据的可用性,<br>实用的存储群集通常会对冗余数据进行条带化<br>通过复制或擦除编码跨多个节点。<br>复制创建相同的数据副本,并由<br>前几代的存储集群,但它带来了大量<br>存储开销,特别是随着当今的巨大增长<br>数据存储。另一方面,擦除编码创建<br>有限的冗余数据通过编码计算,<br>并可证明地保持与<br>与复制相比,存储冗余要少得多[41]。今天的<br>大规模存储集群越来越多地采用擦除编码<br>提供低成本的容错存储(如[2]、[10]、[13],<br>[26],[28]),并报告比<br>复制[13],[26]。<br>在高效存储的同时,擦除编码带来了<br>修理罚款。作为一个例子,我们认为里德所罗门(RS)<br>代码[34],这是一种流行的擦除编码结构<br>用于生产[2],[10],[26],[28]。在高水平上,RS<br>编码将k个数据块编码成n个编码块<br>参数k和n>k<br>块可以重建(或解码)所有原始k数据块。<br>但是,修复丢失的RS代码块需要检索<br>k个可用于解码的块,这意味着两个带宽<br>单个块修复的I/O成本放大了k倍;<br>相反,在复制中,修复丢失的块可以简单地<br>通过检索另一个可用的区块副本完成。<br>高维修罚款是所有时代的基本问题<br>编码结构:维修流量随着存储量的增加而增加<br>冗余减少[8]。因此,有广泛的<br>提高擦除编码修复性能的研究,<br>例如提出理论上证明的擦除码<br>在修复期间最小化修复通信量或I/O(例如,[8],<br>[13] ,[35]),或设计适用于<br>对于所有实用的擦除码,包括RS码(例如,[5],<br>[20] ,[21],[24],[37],[38])。常规维修方法<br>是反应性的,这意味着仅触发维修操作<br>在检测到节点故障后。不过,如果我们能预测<br>即将发生的故障,我们可以提前修复<br>任何即将发生故障的节点的数据丢失以减轻修复<br>任何实际故障发生前的惩罚。<br>最近的研究表明机器学习可以实现<br>生产环境中磁盘故障的准确预测<br>有数以千计的磁盘[6],[18],[23],[42],[43],[45];在一些<br>例,预测精度甚至达到95%[6],<br>[18] ,[23],[45]错误报警率很小。动机<br>高精度磁盘故障预测的潜力,我们可以<br>精确定位即将发生故障(STF)的节点并加速<br>通过耦合两种修复方法修复操作:(i)迁移,<br>其中我们重新定位当前存储的STF块<br>节点到其他健康节点,以及(ii)重建,其中<br>我们通过以下方法重建(或解码)STF节点的块<br>将存储群集中所有正常节点的块检索为<br>在传统的反应性修复方法中。迁移地址<br>固有的带宽和I/O放大问题<br>在擦除编码中,重建利用了<br>所有正常节点的带宽资源。一个悬而未决的问题是<br>如何将迁移与重建紧密结合<br>最大限度地提高修复性能。<br>我们提出了一种快速预测修复方法FastPR<br>小心地将大块的迁移和重建结合起来<br>以最小化为主要目标的STF节点<br>总修复时间。FastPR计划迁移和<br>STF节点块的并行重构<br>时尚,以便利用<br>基础存储群集。我们解决了两种修复方案:<br>零散的修理,储存修理过的STF块<br>跨越存储集群中所有其他现有节点的节点,<br>以及热备份修复,它存储修复后的<br>专用热备用节点中的STF节点。我们提出<br>通过数学分析、大比例尺模拟和Amazon EC2实验,对FastPR进行了深入研究,并对FastPR进行了仿真<br>以下贡献:<br>•我们首先给出了最优<br>在最小化总修复时间方面的预测修复。我们展示<br>
正在翻译中..
 
其它语言
本翻译工具支持: 世界语, 丹麦语, 乌克兰语, 乌兹别克语, 乌尔都语, 亚美尼亚语, 伊博语, 俄语, 保加利亚语, 信德语, 修纳语, 僧伽罗语, 克林贡语, 克罗地亚语, 冰岛语, 加利西亚语, 加泰罗尼亚语, 匈牙利语, 南非祖鲁语, 南非科萨语, 卡纳达语, 卢旺达语, 卢森堡语, 印地语, 印尼巽他语, 印尼爪哇语, 印尼语, 古吉拉特语, 吉尔吉斯语, 哈萨克语, 土库曼语, 土耳其语, 塔吉克语, 塞尔维亚语, 塞索托语, 夏威夷语, 奥利亚语, 威尔士语, 孟加拉语, 宿务语, 尼泊尔语, 巴斯克语, 布尔语(南非荷兰语), 希伯来语, 希腊语, 库尔德语, 弗里西语, 德语, 意大利语, 意第绪语, 拉丁语, 拉脱维亚语, 挪威语, 捷克语, 斯洛伐克语, 斯洛文尼亚语, 斯瓦希里语, 旁遮普语, 日语, 普什图语, 格鲁吉亚语, 毛利语, 法语, 波兰语, 波斯尼亚语, 波斯语, 泰卢固语, 泰米尔语, 泰语, 海地克里奥尔语, 爱尔兰语, 爱沙尼亚语, 瑞典语, 白俄罗斯语, 科西嘉语, 立陶宛语, 简体中文, 索马里语, 繁体中文, 约鲁巴语, 维吾尔语, 缅甸语, 罗马尼亚语, 老挝语, 自动识别, 芬兰语, 苏格兰盖尔语, 苗语, 英语, 荷兰语, 菲律宾语, 萨摩亚语, 葡萄牙语, 蒙古语, 西班牙语, 豪萨语, 越南语, 阿塞拜疆语, 阿姆哈拉语, 阿尔巴尼亚语, 阿拉伯语, 鞑靼语, 韩语, 马其顿语, 马尔加什语, 马拉地语, 马拉雅拉姆语, 马来语, 马耳他语, 高棉语, 齐切瓦语, 等语言的翻译.

Copyright ©2024 I Love Translation. All reserved.

E-mail: