[论文解读] FAST CLOUD: Pushing the Envelope on Delay Performance of Cloud Storage with Coding

Guanfeng Liang, Ulaş C. Kozat|arXiv (Cornell University)|Jan 7, 2013

Advanced Data Storage Technologies参考文献 15被引用 6

一句话总结

该论文提出 FAST CLOUD，一种负载自适应的编码与分块框架，通过结合纠删码、并行连接和动态码率选择，显著降低了云存储访问延迟。该方法在仅 1.75 倍存储开销下，将 2 MB 文件的 99 百分位延迟降低高达 85%，通过基于实时队列积压阈值的队列感知自适应机制，优于固定码率策略。

ABSTRACT

Our paper presents solutions that can significantly improve the delay performance of putting and retrieving data in and out of cloud storage. We first focus on measuring the delay performance of a very popular cloud storage service Amazon S3. We establish that there is significant randomness in service times for reading and writing small and medium size objects when assigned distinct keys. We further demonstrate that using erasure coding, parallel connections to storage cloud and limited chunking (i.e., dividing the object into a few smaller objects) together pushes the envelope on service time distributions significantly (e.g., 76%, 80%, and 85% reductions in mean, 90th, and 99th percentiles for 2 Mbyte files) at the expense of additional storage (e.g., 1.75x). However, chunking and erasure coding increase the load and hence the queuing delays while reducing the supportable rate region in number of requests per second per node. Thus, in the second part of our paper we focus on analyzing the delay performance when chunking, FEC, and parallel connections are used together. Based on this analysis, we develop load adaptive algorithms that can pick the best code rate on a per request basis by using off-line computed queue backlog thresholds. The solutions work with homogeneous services with fixed object sizes, chunk sizes, operation type (e.g., read or write) as well as heterogeneous services with mixture of object sizes, chunk sizes, and operation types. We also present a simple greedy solution that opportunistically uses idle connections and picks the erasure coding rate accordingly on the fly. Both backlog and greedy solutions support the full rate region and provide best mean delay performance when compared to the best fixed coding rate policy. Our evaluations show that backlog based solutions achieve better delay performance at higher percentile values than the greedy solution.

研究动机与目标

• 为解决 Amazon S3 等云存储系统中因小到中等大小对象导致的高波动性和差延迟性能问题。
• 分析分块、纠删码和并行访问对云存储中排队和处理延迟的影响。
• 设计负载自适应算法，基于队列积压阈值动态选择码率，以最小化端到端延迟。
• 支持异构工作负载，涵盖混合对象大小、分块大小和操作类型（读/写）。
• 开发一种贪婪的实时自适应策略，机会性地利用空闲连接，并实时调整编码速率。

提出的方法

• 提出一种排队模型，将 M/G/k 拓展为包含纠删码引起的相互依赖服务时间的模型，其中任意 k 个编码片段中的完成即足以完成请求。
• 推导出在分块、前向纠错（FEC）和多服务器环境下，考虑负载相关服务速率的平均延迟的闭式近似解。
• 提出一种负载自适应算法，通过预计算的队列积压阈值为每个请求选择最优码率，以最小化延迟。
• 设计一种贪婪的在线算法，基于可用空闲连接动态调整编码速率，无需预先知晓负载情况。
• 采用凸优化框架建模延迟与负载之间的权衡，目标函数同时包含排队延迟和服务延迟。
• 建立理论条件，证明最优码向量位于可行区域内，确保自适应策略的收敛性和稳定性。

实验结果

研究问题

• RQ1 在 Amazon S3 等云存储系统中，纠删码结合分块和并行访问如何影响延迟分布？
• RQ2 码率和分块对云存储系统中可支持的请求速率和排队延迟有何影响？
• RQ3 基于实时队列积压的动态码率自适应是否能超越固定码率编码策略，提升延迟性能？
• RQ4 在混合对象大小、分块大小和操作类型（读/写）的异构工作负载下，系统性能如何？
• RQ5 在云存储中使用自适应纠删码时，存储开销与延迟降低之间的权衡是什么？

主要发现

• 对于 2 MB 文件，FAST CLOUD 相较于未编码存储，将平均延迟降低 76%，第 90 百分位延迟降低 80%，第 99 百分位延迟降低 85%。
• 所提出的基于队列积压的负载自适应算法在高百分位延迟上优于贪婪在线方法。
• 系统在高负载下仍能支持未编码存储的完整速率区域，同时保持更低的平均延迟。
• 只有在结合负载感知码率自适应时，纠删码与分块才能改善延迟性能；否则，由于负载增加，性能反而下降。
• 最优码率并非固定，而是取决于当前负载和队列积压情况，较低码率可在高利用率下扩展系统容量。
• 理论分析证实，性能优良的码向量可被完全排序，从而实现实时高效搜索与自适应。