Italian for Beginners: The Next Steps for SLO-Based Management

---------------------------------------------------

摘要：

Literature is rife with compelling ideas on simplifying storage management using service-level objectives (SLOs). However, very few of these ideas have actually been productized, despite the fact that many of the original ideas came from industry and were developed more than a decade ago. While many good research ideas do not become products, in this case, we believe that there are important reasons why adoption has been slow. This paper investigates the reasons for slow adoption and discusses ideas that can truly simplify storage management using SLOs.

研究背景：

Recent research [19] has extended and explored automated management using service-level objectives (SLOs) [17, 30, 36] to meet application goals in the cloud while keeping management costs low.

With SLOs, administrators can focus more on what they need from storage than on how it is achieved.

[17] Kephart and Chess. The Vision of Autonomic Computing. IEEE Computer, 36(1), 2003.

[30] Strunk et al. Using Utility Functions to Provision Storage Systems. In FAST' 08.

[36] Wilkes. Traveling to Rome: A Retrospective on the Journey. In R2D2' 08.

现象：However, very few of these ideas have been productized, especially in the context of storage systems. Unless we first address productization, we may not be solving the real problems, whether for the cloud or elsewhere.

问题：

Our analysis shows that (a) SLOs are not as simple to specify as we would like, (b) system models, which proactively assess system behavior, have considerable error relative to manual approaches, and (c) the cost of remedying a modeling error is too high.

为了解决上述问题，建议研究方向：

In order to address these issues, we offer the following research directions:

l         We should focus on process, not product; processes that limit the scope of SLO-based management to a pre-defined, well-known set of SLOs (“qualified SLOs”) will greatly improve (i) the ability of users to select SLOs, (ii) the accuracy of models, and (iii) the ability of vendors to support systems.

l         We should leverage the similarity of workload instances across the data center or even the entire customer base to gather data on workloads and continually update system models and support workflows.

l         We should develop lightweight dynamic reconfiguration techniques to mitigate the cost of modeling errors.

注1：为什么叫这样的题目？

The title is a nod to research by Wilkes et al. on SLO-based management for storage systems, summarized as “Traveling to Rome” [36]. Our paper is about the important next steps.

[36] Wilkes. Traveling to Rome: A Retrospective on the Journey. In R2D2’08.

注2：Coterie Join Algorithm

Given a set of nodes in a distributed system, a coterie is a collection of subsets of the set of nodes such that any two subsets have a nonempty intersection and are not properly contained in one another. A subset of nodes in a coterie is called a quorum.

参考：http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.54.4687