OctoFAS: A Two-Level Fair Scheduler That Increases Fairness in Network-Based Key-Value Storage
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis research paper proposes a two-level fair I/O scheduling mechanism to solve the fairness problem in I/O services from multi-tenants in a multi-tenant environment (inherent to, for instance, the network-based key-value storage system with Intel SPDK). The results support that the proposed methods achieve improved fairness by approximately 10% without significant throughput degradation.
However, the introductory and closing sections must be improved, and the grammar in English language should be revised.
1. What is the main question addressed by the research?
This research paper proposes a two-level fair I/O scheduling mechanism to solve the fairness problem in I/O services from multi-tenants in a multi-tenant environment (inherent to, for instance, the network-based key-value storage system with Intel SPDK). The results support that the proposed methods achieve improved fairness by approximately 10% without significant throughput degradation.
2. Do you consider the topic original or relevant in the field? Does it address a specific gap in the field?
The topic is not original as it stands, for the fairness problem it addresses has been studied in several of the references already cited in the manuscript (see [5-12]). However, only a handful of works deal with the concept of disaggregated storage (see [13]).
3. What does it add to the subject area compared with other published material?
This paper introduces an alternative to the network-connected key-value store studied in [13] to solve the multi-tenancy problem. The authors claim that such an alternative enhances the overall system performance and fairness simultaneously.
4. What specific improvements should the authors consider regarding the methodology? What further controls should be considered?
The methodology is sound and the presentation is well organized. However, the authors should include execution time analyses of algorithms 1-4 and explain more thoroughly their choices for the parameters they set in their evaluation setup.
5. Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed?
The conclusions are consistent with the results, but that section should be revised, for the sentence in lines 592-593 seems out of place.
6. Are the references appropriate?
They are. However, the authors should provide at least one more reference to exemplify the study of the concept of disaggregated storage.
7. Please include any additional comments on the tables and figures.
The tables and figures are beautifully designed, but the chromatic scale in Figure 4 does not contribute to the readability and interpretation of the relation between I/O Latency and Core utilization. I suggest adding numeric values to the graph.
The quality of the usage of grammar and English language in general should be improved. Reading the paper is challenging due to this reason.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors This paper tackles a fairness problem in a network-based KV storage using Intel SPDK in a multi-tenant environment. The proposed OctoFAS two-level scheduler combines inter-core and intra-core scheduling to improve throughput and fairness between tenants. - In 2.1 (line 97), the DBEV acronym is not explicitly defined - Consider replacing the plots in Figure 4 with boxplots to illustrate the results, including the observed distribution, since a lot of points seem to overlap, and the scale does not help with that. - Paper is well motivated with clear argumentation and supporting experiments to demonstrate the issue in practice. - In 4.1 (lines 284-286), it is unclear how changing the order of I/O processing can be considered a "new I/O event scheduling technique." In terms of I/O, request reordering is not novel. - In 4.2.1 (lines 315-317), it is unclear how and why the "correction" is applied based on the network time. - In 4.3.1 (line 349), the authors state that the inter-core scheduling they propose operates similarly to the previous SPDK-based KVS. How do they compare in terms of performance? - In 4.3.2 (lines 393 and 407), what are the guidelines/recommendations to identify the alpha constant and backwarding iterations suitable for each system? Is this via exploratory runs, or are there some guidelines that could be used? - In 4.4.2 (lines 435-437), are those changes available somewhere? It would be good to add a reference to a permanent repository to promote the usage of this solution. - In 4.4.3 (line 442), the number 6 is missing a space after Figure. - In 5.1, why was the random workload used? Do the results and behavior hold for the other workloads?- The last sentence of the conclusion seems incomplete and repeated from the previous one (which also does not seem finished).Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper presents OctoFAS, a scheduling system for enhancing fairness in Intel SPDK-based key-value storage systems, especially in multi-tenant environments. It addresses the issue of unfair I/O resource allocation among tenants, a problem overlooked by SPDK despite its throughput optimization by introducing a dual-layered fairness scheduling mechanism, comprising inter-core and intra-core scheduling, to balance load and prioritize under-resourced tenants. The system's effectiveness is validated through evaluations, demonstrating a significant improvement in fairness while maintaining overall throughput. This innovation is crucial for high-performance storage systems. Overall, the paper is in high-quality.
The suggestions are stated as follows:
1. L86, an example of this ultra-low latency is preferred.
2. For figure2(b), L182 – 188, please add a reference or explain some details of this migration.
3. L227, the throughput values seem not to be constant in figure 3(b)
4. L442, spacing issue “figure6”
5. L539, where is the (e) and (f) in figure 9?
6. L591-592, indent issue.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe questions and concerns raised by this referee have been adequately addressed. I therefore recommend the publication of the paper.
