Familiarity Recommendation Some knowledge (I am marginally aware of research work in this topic) (2) Likely accept (top 20% but not top 10%, significant contribution) (4) Contributions (What are the major issues addressed in the paper? Do you consider them important? Comment on the novelty, creativity, impact, and technical depth in the paper.) The paper presents a capacity allocation algorithm for cloud-based services, e.g., Amazon’s Simple Storage Service (S3) or Elastic Compute Cloud (EC2). The considered cloud consists of n servers which can adjust their capacity C_i (i=1..n) dynamically. Each server is faced with a fixed demand D_i and the customer experiences a performance level q_i. This problem is called distributed rate limiting. The proposed algorithm assigns capacities to the n servers incrementally and in a distributed way such that all servers provide the same performance level. Further contributions of this paper are a proof that the algorithm converges and a simulative evaluation. Strengths (What are the major reasons to accept the paper?) The paper significantly extends the related work which only describes distributed rate limiting (DLR) algorithms for specific problems. In contrast, this paper presents a generalized DLR algorithm. Furthermore, a proof is presented how to choose parameters such that the algorithm converts. The theoretical investigation is validated via simulation experiments. Weaknesses (What are the major reasons NOT to accept the paper?) There are only some minor issues compared to the positives: The motivation of DRL was not clear to me, I had to extract it from the related work. Some English problems (see below) and some problems in the presentation, but they can be easily corrected. Detailed Comments (Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors.) - It would be good to provide a more specific example for DLR in the introduction so that the reader can understand for which purpose it can be used and show the differences to load balancing - Sect. I: “… to evaluate the performance through experienced latency; etc.” -> “… latency, etc.” - Sect. I-A: “Let a cloud-based …” -> “A cloud-based …” - Sect. I-A: footnote 4: Is the content necessary for the paper? If not, I’d remove the footnote - Sect. I-A: “The dynamical system that … are nonlinear and implicit” -> “The system … is nonlinear and implicit” - Sect. III-A: “… the function f_i:x->x-\lambda_i” -> “… f:x->\lambda_i-x” - Descriptions of Fig. 3, 4, and 6 should contain whether dynamic or static demands are used - It should be indicated in Fig. 3, 5, and 6 which color corresponds to which q_i - The dimension of the x-axis is missing in Fig. 3-6 - Sect. IV: “… and that that information” -> “… and that information” - Setc. IV: “what is relevant performance indicator” -> “what is a relevant performance indicator” Review 2 Familiarity Recommendation Familiar (I am well aware of research work in this topic) (3) Definite accept (top 10%, excellent paper) (5) Contributions (What are the major issues addressed in the paper? Do you consider them important? Comment on the novelty, creativity, impact, and technical depth in the paper.) The paper represents an interesting generalization of prior work on distributed rate limiting. While the paper is focused upon a theoretical treatment of the subject, it sets the foundation for subsequent theoretical research in the area of distributed rate limiting and for a bevy of domain-specific proposals that put that theory to the test in real systems. Strengths (What are the major reasons to accept the paper?) The concept of a unified, theoretical distributed rate limiting framework is novel and interesting, and has the potential to spur substantial future work. Weaknesses (What are the major reasons NOT to accept the paper?) The writing is stilted in parts - the authors should carefully edit for logical flow and for grammar. The experimental evaluation in Section III is overly simplistic. Detailed Comments (Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors.) At a high level the work could be improved the most by working on its presentation - the writing and layout of content. In addition, it is difficult to read and interpret many of the figures in the paper, so that could benefit from addition clarification as well (and the use of a vector format like EPS to ensure that graphs are not distorted by scaling). I'm okay with you selecting arbitrary graphs and communication models in Section III, but it would be helpful if you were to thoroughly justify your decision to use each. You should refer to the weaknesses described in Section IV much earlier in the paper - many of these are fundamental to a theory of fully general distributed rate limiting systems, not to mention their actual implementation. The ending summary section is valuable. If possible, you should expand this a bit, to add more discussion of the open questions you introduce and to add more discussion of your results rather than focusing mainly on the context of the work. Review 3 Familiarity Recommendation Some knowledge (I am marginally aware of research work in this topic) (2) Definite accept (top 10%, excellent paper) (5) Contributions (What are the major issues addressed in the paper? Do you consider them important? Comment on the novelty, creativity, impact, and technical depth in the paper.) This paper provides a generalized mechanism for distributed rate limiting so that performance metrics for a particular service can be equalized across group of servers. The work does not assume TCP as with previous work. It provides high quality discussions, well formulated proofs of convergence, examples showing the effectiveness of the algorithms with various loading schemes and varies types of loading changes (step, gradual) within the simulation time. Strengths (What are the major reasons to accept the paper?) Well formulated algorithms and proofs, good demonstrations of the effectiveness through simulations. As a paper, the writing is outstanding, readable, and clear. Weaknesses (What are the major reasons NOT to accept the paper?) The paper assumes convex, differentiable functions, but this is not so much a weakness since such assumptions are reasonable. Otherwise, this reviewer cannot identify any other significant weaknesses. Detailed Comments (Please provide detailed comments that will help the TPC assess the paper and help provide feedback to the authors.) This reviewers sees no substantive comments that need to be made about the paper.