Dynamo: Amazon's Highly Available Key-value Store

• p. 2 Introduction

  In the past year, Dynamo has been the underlying storage technology

• p. 14 Section 6.6 Discussion

  Many Amazon internal services have used Dynamo for the past two years

• p. 15 Section 7 Conclusions

  The production use of Dynamo for the past year demonstrates

• p. 2 Introduction, paragraph 2

  In the past year, Dynamo has been the underlying storage

Only the authors can say what is right here.

Judge's note: Checked all three passages in the images of pp.2, 14 and 15. The Introduction might refer only to core e-commerce services, but the Conclusions speak of production use in general and say one year, while 6.6 says two years. The paper does not reconcile the two periods, so only the authors can say which is right.

Second check: another model agreed

• p. 11 Section 6.1

  lowering the 99.9th percentile latency by a factor of 5 during peak traffic

• p. 10 Figure 5

  direct BDB writes

• p. 10 Figure 5

  Comparison of performance of 99.9th percentile latencies for buffered vs. non-buffered writes

Only the authors can say what is right here.

Judge's note: Read Figure 5 on p.10 by eye. The direct BDB writes reach about 180 ms at their highest spikes. Buffered writes stay around 55-70 ms through the peak period and are about 40 ms off-peak. A ratio of about 5 is not visible in the plot unless an isolated spike is compared with an off-peak buffered value. Because the values are read by eye, this is a query to the authors: what is the basis for 'factor of 5'?

Second check: another model agreed

• p. 11 Section 6.1

  during a period of 30 days

• p. 10 Figure 4 caption

  The intervals between consecutive ticks in the x-axis correspond to 12 hours

Only the authors can say what is right here.

Judge's note: Corrected description. In the Figure 4 crop I count roughly 29-30 tick intervals across the x-axis, and the average-read line shows about 15 daily cycles. At 12 h per tick that is about 15 days, not 30. So either the text's '30 days' or the caption's tick spacing looks inconsistent. The tick count is approximate, so this goes to the authors. The reviewer's other points do not hold. The 'Dec. 2006' caption is not in conflict with the text. The 99.9th-percentile write line reads about 250 ms on the log scale, which is reasonably 'around 200 ms'.

Second check: another model agreed

• p. 1 Categories and Subject Descriptors

  D.4.2 [Operating Systems]: Performance;

• p. 1 Categories and Subject Descriptors

  D.4.2 [Operating Systems]: Performance

• p. 1

  D.4.2 [Operating Systems]: Storage Management; D.4.5 [Operating Systems]: Reliability; D.4.2 [Operating Systems]: Performance;

Only the authors can say what is right here.

Judge's note: Rests on outside knowledge: in the ACM CCS 1998, Operating Systems/Performance is D.4.8. The page image (p.1) shows D.4.2 attached to both Storage Management and Performance. One code cannot stand for two different descriptors, so one of them is wrong. The correct code comes from outside the files.

Second check: none (judge only)

• p. 1 Section 1, first paragraph

  In addition, to support continuous growth, the platform needs to be highly scalable.

Judge's note: The page 1 image shows both sentences in the same paragraph with nearly identical wording. The repetition is redundant.

Second check: none (judge only)

• p. 1 Section 1, first sentence

  tens of millions customers

• p. 1 Introduction, first sentence

  serves tens of millions customers

• p. 1 1. Introduction, para 1

  serves tens of millions customers at peak times

Judge's note: Seen in the page 1 image, first sentence of the Introduction.

Second check: none (judge only)

• p. 2 Section 1

  served tens of millions requests

Judge's note: Seen in the page 2 image, Introduction paragraph on the Shopping Cart Service.

Second check: none (judge only)

• p. 4 3.2

  resolves updated

Judge's note: Seen in the page 4 image, Section 3.2. The word form is wrong.

Second check: none (judge only)

• p. 5 Figure 2 annotation

  range (A,B)

• p. 6 Section 4.3

  (A, B], (B, C], and (C, D]

Judge's note: The images of pp.5–6 confirm both notations. This looks like informal notation in the figure, but it does not match the text's interval convention.

Second check: none (judge only)

• p. 6 Section 4.4, first paragraph

  updates.. If there are no failures

Judge's note: The page 6 image shows two periods.

Second check: none (judge only)

• p. 6 Section 4.2, first paragraph

  The principle advantage of consistent hashing

• p. 11 Section 6.1

  Dynamo’s principle design goal

Judge's note: Both instances are visible in the images of pp.6 and 11. Section 4.7 spells it correctly: "The principal advantage of Merkle tree".

Second check: none (judge only)

• p. 6 Section 4.2, third bullet

  that a node is responsible can decided based on its capacity

• p. 6 4.2 bullets

  The number of virtual nodes that a node is responsible can decided based on its capacity

Judge's note: Seen in the page 6 image, third bullet of Section 4.2.

Second check: none (judge only)

• p. 6 Section 4.4, vector clocks paragraph

  by examine their vector clocks

Judge's note: Seen in the page 6 image, Section 4.4.

Second check: none (judge only)

• p. 7 Figure 3

  D5 ([Sx,3],[Sy,1][Sz,1])

Judge's note: The page 7 image shows the missing comma. All other clocks in the figure separate entries with commas, and the text gives [(Sx, 3), (Sy, 1), (Sz, 1)].

Second check: none (judge only)

• p. 8 Section 4.6

  durably written the key it to its local store

Judge's note: Seen in the page 8 image, Section 4.6. The word 'it' is redundant.

Second check: none (judge only)

• p. 9 Section 4.9

  some existing nodes no longer have to some of their keys

Judge's note: The crop of p.9 shows the text exactly as quoted: 'no longer have / to some of their keys'. A verb is missing.

Second check: none (judge only)

• p. 10 Figure 4 x-axis label

  peak seson in Dec. 2006

• p. 10 Figure 4, x-axis label

  (hourly plot of latencies during our peak seson in Dec. 2006)

Judge's note: Page 10 image and crop C021-b both show 'seson' in the Figure 4 axis label. It is part of the figure graphic, which is why the text layer does not have it.

Second check: none (judge only)

• p. 13 Figure 8 y-axis label

  Efficieny (mean load/max load)

• p. 10 Figure 4 x-axis label

  peak seson in Dec. 2006

Judge's note: The 'Efficieny' label is visible on the p.13 figure image. The 'seson' item also listed here is the same defect as C017.

Second check: none (judge only)

• p. 11 Section 6.2

  by a value a less than a certain threshold

• p. 11

  by a value a less than a certain threshold (here 15%)

Judge's note: The p.11 image shows 'by a value a less than a certain threshold (here 15%)'.

Second check: none (judge only)

• p. 11 6.1

  Dynamo’s principle design goal

Judge's note: The p.11 image shows 'principle design goal'. It is a misspelling of 'principal'.

Second check: none (judge only)

• p. 14 Section 6.6

  based on their needs..

Judge's note: The double period is visible in the p.14 image.

Second check: none (judge only)

• p. 15 Reference [15]

  Berkeley, CA, 12-12..

• p. 15 Ref [14]

  Beehive: O(1)lookup

Judge's note: The p.15 crops show both defects as quoted.

Second check: none (judge only)

• p. 15 7. Conclusions

  an eventual-consistent storage system

Judge's note: The p.15 image shows 'eventual-consistent'. The text layer shows 'eventually-consistent' in the introduction (line 125) and 'eventually consistent' elsewhere. This is a cosmetic inconsistency.

Second check: none (judge only)