Cassandra: A Decentralized Structured Storage System
ACM SIGOPS Operating Systems Review, 2010 · paper · Computer science
Describes Cassandra, Facebook's distributed storage system for large structured data on commodity servers, with no single point of failure. It offers a simple data model rather than full relational support, and handles high write throughput.
Checked: PDF from cs.cornell.edu, fetched 4 Oct 2026. Page numbers are those of that PDF.
4 findings
Reference or label Section 5.2: "As is explained in Section 5.1 every node is aware of every other node in the system" points to Section 5.1 (Partitioning), which does not explain this; the material is in Sections 5.3 and 5.4.
- p. 3 Section 5.2 Replication, second paragraph
As is explained in Section 5.1 every node is aware of every
- p. 2 Section 5.1 Partitioning
5.1 Partitioning
- p. 3 Section 5.2 Replication, second paragraph
As is explained in Section 5.1 every node is aware of every other node in the system
- p. 3 Section 5.4 Bootstrapping
This is how we know about all nodes and their respective po- sitions in the ring.
Only the authors can say what is right here.
Judge's note: Read all of Section 5.1 on the page 2 image. It covers consistent hashing, the coordinator, and load balancing, and says nothing about nodes knowing each other. Gossip-based membership is in 5.3, and 5.4 says "This is how we know about all nodes and their respective positions in the ring". Section 2 makes the every-node-knows-every-node statement only about Dynamo. The intended target is probably 5.3 or 5.4, but only the authors can say which.
Second check: another model agreed
Report a problem with this finding
Reference or label Section 5.2 cites itself for content it does not contain: "by relaxing the quorum requirements as described in Section5.2."
- p. 3 Section 5.2 Replication, second paragraph
laxing the quorum requirements as described in Section5.2.
- p. 3 Section 5.2 Replication, second paragraph
laxing the quorum requirements as described in Section5.2
- p. 3 Sec 5.2, last paragraph
relaxing the quorum requirements as described in Section5.2
- p. 3 Section 5.2
as described in Section5.2
Only the authors can say what is right here.
Judge's note: Page 3 image: the sentence sits in the second paragraph of Section 5.2. Neither Section 5.2 nor any other section describes relaxing quorum requirements. A text search finds quorum only in the Section 5 overview (waits for a quorum of replicas or responses) and in Section 5.7 (wait for a quorum of responses). The pointer is self-referential and has no target; the intended one is unclear. The missing space in "Section5.2" is handled under C009.
Second check: another model agreed
Report a problem with this finding
Reference or label Reference [12] (Massie, Chun and Culler, Ganglia) reads "Parallel Computing, 30:2004, 2004.": the year appears in the page field after the volume, and no issue or page range is given.
- p. 6 References, entry [12]
Computing, 30:2004, 2004.
- p. 6 Reference [12]
Parallel Computing, 30:2004, 2004.
- p. 6
ience. Parallel Computing, 30:2004, 2004. [13] Benjamin Reed
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the true issue and page range of the Ganglia article in Parallel Computing vol. 30, which are needed for the fix and are not in the files. The defect itself is visible on the page 6 image: "30:2004, 2004" repeats the year where the pages should be, and a single page 2004 in volume 30 is implausible.
Second check: another model agreed, code check
Report a problem with this finding
Wording Section 7 states "We have empirically demonstrated that Cassandra can support a very high update throughput while delivering low latency", but the paper reports no write or update throughput measurement; Section 6.1 says "We show some production measured numbers for read performance" and the only table gives read latencies.
- p. 5 Section 7 Conclusion
We have empirically demonstrated that Cassandra can support a very high update throughput while deliver- ing low latency
- p. 5 Section 6.1, sentence before the latency table
We show some production measured numbers for read perfor- mance.
Only the authors can say what is right here.
Judge's note: Searched the full text for "throughput" and for write figures. The hits are requirements or design statements: "billions of writes per day" in the introduction, asynchronous replication, sequential writes. The only measurements are the read-latency table (min, median, max), the 40 MB/sec bootstrap transfer rate, failure-detection times, and 50+TB on 150 nodes. The conclusion's wording therefore goes beyond the data presented. The second reviewer note (the latency table has no number or caption, and its columns are in the opposite order to the text) is a style point and is not confirmed as a defect.
Second check: another model agreed
Report a problem with this finding
26 minor (typos and the like; most checked by the judge only)
Typo Introduction: "serves hundreds of millions users at peak times" is missing "of" (should be "hundreds of millions of users").
- p. 1 Section 1 Introduction, first sentence
serves hundreds of millions users at peak times using tens of
- p. 1 Section 1 Introduction, first sentence
serves hundreds of millions users at peak times
- p. 1 Introduction
hundreds of millions users at peak times
Judge's note: Read on the page 1 image, Section 1, first sentence. The wording is printed as quoted, so it is not an extraction artefact.
Second check: none (judge only)
Report a problem with this finding
Typo Copyright block still carries the unfilled ACM template placeholder: "Copyright 200X ACM X-XXXXX-XX-X/XX/XX ...$10.00."
- p. 1 Copyright block, foot of left column
Copyright 200X ACM X-XXXXX-XX-X/XX/XX ...$10.00.
Judge's note: Visible at the foot of the left column on the page 1 image. It is a template leftover, common in workshop author versions; cosmetic only.
Second check: none (judge only)
Report a problem with this finding
Typo Section 2 (Related Work), Dynamo paragraph: the printed text "This is can be very limiting in environments where systems need to handle a very high write throughput" contains a stray "is"; it should read "This can be very limiting".
- p. 2 Section 2 Related Work, Dynamo paragraph
This is can be very limiting in environments where systems need
- p. 2
This is can be very limiting in environments
- p. 2 Sec 2
This is can be very limiting
Judge's note: I checked the crop of page 2, left column, Dynamo paragraph. "This is" ends one line and "can be very limiting" begins the next, and both words are printed on the page. So this is not an extraction artefact. It is a cosmetic grammatical slip.
Second check: another model agreed
Report a problem with this finding
Typo Section 2: "Dynamo detects updated conflicts using a vector clock scheme" should read "update conflicts".
- p. 2 Section 2 Related Work, Dynamo paragraph
Dynamo detects updated conflicts using a vector clock
- p. 2 Section 2
Dynamo detects updated conflicts
Judge's note: Page 2 image shows "updated conflicts". The same section uses "Update conflicts are typically managed" (page 1) and "resolves update conflicts" (page 2), so this is a typo.
Second check: none (judge only)
Report a problem with this finding
Typo Section 3: "Cassandra exposes two kinds of columns families, Simple and Super column families" should read "column families".
- p. 2 Section 3 Data Model
Cassandra exposes two kinds of columns families,
- p. 2 Section 3
two kinds of columns families
Judge's note: Page 2 image shows "columns families". The same sentence and the surrounding text use "column families".
Second check: none (judge only)
Report a problem with this finding
Typo Section 5.1: "The application specifies this key and the Cassandra uses it to route requests" has a stray article before "Cassandra".
- p. 2 Section 5.1
the Cassandra uses it to route requests
Judge's note: Page 2 image confirms the wording. Elsewhere "the Cassandra" is always followed by a noun (system, cluster, instance, process); here it stands alone, so "the" is stray or a noun such as "system" is missing.
Second check: none (judge only)
Report a problem with this finding
Typo Section 5.2: missing space in "as described in Section5.2."
- p. 3 Section 5.2, second paragraph
Section5.2
Judge's note: Page 3 image shows "Section5.2" printed fused, while "Section 5.1" three lines above has the space. This is the typographic defect only; the wrong pointer target is C008.
Second check: none (judge only)
Report a problem with this finding
Typo Section 5.4: "to a node trying to join a wrong Cassandra instance it can thwarted based on the cluster name" is missing "be" (should be "it can be thwarted").
- p. 3 Section 5.4 Bootstrapping, second paragraph
to a node trying to join a wrong Cassandra instance it can
Judge's note: Page 3 image, right column: "it can" ends one line and "thwarted based on the cluster name" starts the next, with no "be".
Second check: none (judge only)
Report a problem with this finding
Typo Section 5.2: "The metadata about the ranges a node is responsible is cached locally" is missing "for" (should be "responsible for is cached").
- p. 3 Section 5.2 Replication
metadata about the ranges a node is responsible is cached
- p. 3 Section 5.2
ranges a node is responsible is cached
Judge's note: Page 3 image confirms the wording. The next sentence has the correct form: "knows what ranges it was responsible for".
Second check: none (judge only)
Report a problem with this finding
Typo Section 5.2: "contact the leader who tells them for what ranges they are replicas for and leader makes a concerted effort" has a duplicated "for" and no article before "leader".
- p. 3 Section 5.2 Replication
them for what ranges they are replicas for and leader makes
Judge's note: Page 3 image confirms the wording. Grammar slip only; the meaning is clear.
Second check: none (judge only)
Report a problem with this finding
Typo Page 3, right column, the paragraph on manual configuration errors and cluster names: the printed text "to a node trying to join a wrong Cassandra instance it can thwarted based on the cluster name" is missing "be"; it should read "it can be thwarted".
- p. 3
it can thwarted based on the cluster name.
- p. 3 Sec 5.4
it can thwarted based on the cluster name
Judge's note: I checked the crop of page 3, right column. "it can" ends one line and "thwarted based on the cluster name" begins the next, with no "be" on either line. This is a missing word, which is cosmetic.
Second check: another model agreed
Report a problem with this finding
Typo The failure-detector value is the symbol Φ in Section 5.3.1 ("This value is defined as Φ.") but is spelled "PHI" in Section 6 ("a slightly conservative value of PHI, set to 5"); Section 5.3.1 also uses Φ for both the emitted value and the threshold ("Given some threshold Φ ... when Φ = 1").
- p. 5 Section 6, third bullet
of PHI, set to 5, the average time to detect failures in
- p. 3 Section 5.3.1
1% with Φ = 2, 0.1% with Φ = 3, and so on.
- p. 5 Section 6, third bullet
of PHI, set to 5
- p. 3 Section 5.3.1 Failure Detection
This value is defined as Φ.
Judge's note: Checked both places on the page 3 and page 5 images. The notation is inconsistent but the meaning is recoverable, and "PHI" may be the name of the configuration parameter, so either form could be legitimate. The double use of Φ for value and threshold follows the phrasing of the cited accrual-detector paper. Cosmetic only.
Second check: none (judge only)
Report a problem with this finding
Typo Section 5.7: "The Cassandra process on a single machine is primarily consists of the following abstractions" has a superfluous "is".
- p. 4 Section 5.7 Implementation Details, first sentence
The Cassandra process on a single machine is primarily
- p. 4
The Cassandra process on a single machine is primarily consists of
Judge's note: Page 4 image, first sentence of Section 5.7: "is primarily" ends one line and "consists of" starts the next. Both verbs are printed.
Second check: another model agreed
Report a problem with this finding
Typo Section 5.7: "In Cassandra we use a rolling a commit log where a new commit log is rolled out after an older one exceeds a particular, configurable, size" has a duplicated article; it should read "a rolling commit log".
- p. 4 Section 5.7, commit log paragraph
a commit log where a new commit log is rolled out after an
Judge's note: Read on the page 4 image, right column: "rolling" ends one line and "a commit log" starts the next, so the extra "a" is in the typeset text and is not a line-break or extraction artefact.
Second check: none (judge only)
Report a problem with this finding
Typo Page 4, right column, the paragraph on purging commit log entries: the printed text "In Cassandra we use a rolling a commit log where a new commit log is rolled out" repeats the article "a"; it should read "a rolling commit log".
- p. 4
we use a rolling a commit log where a new
- p. 4 Section 5.7
we use a rolling a commit log
Judge's note: I checked the crop of page 4, right column. "we use a rolling" ends one line and "a commit log where a new commit log is rolled out" begins the next, so the article "a" is printed twice. This is cosmetic.
Second check: another model agreed
Report a problem with this finding
Typo Section 5.7: "The Cassandra process on a single machine is primarily consists of the following abstractions" has a stray "is"; it should read "primarily consists of".
- p. 4 Section 5.7
is primarily consists of the following abstractions
Judge's note: Read on the page 4 image, left column, first sentence of Section 5.7. The text is printed as quoted.
Second check: none (judge only)
Report a problem with this finding
Typo Section 6, third bullet: "In one particular experiment in a cluster of 100 nodes time to taken to detect a failed node was in the order of two minutes" has a stray "to"; it should read "the time taken to detect".
- p. 5 Section 6, third bullet
cluster of 100 nodes time to taken to detect a failed
- p. 4 Sec 5.7 first sentence
is primarily consists of the following abstractions
- p. 5 Sec 6, failure detector bullet
time to taken to detect a failed
Judge's note: Read on the page 5 image, left column, failure-detector bullet. The text is printed as quoted. One reviewer note in this cluster also raises "is primarily consists of" on page 4; that is real but is covered by C018.
Second check: none (judge only)
Report a problem with this finding
Typo Section 6, second bullet: "there have been some applications that have asked for transactional mainly for the purpose of maintaining secondary indices" is missing a noun after "transactional" (for example "semantics" or "support").
- p. 5 Section 6, second bullet
cations that have asked for transactional mainly for the
Judge's note: Read on the page 5 image, left column. No word follows "transactional" in the typeset text, so this is not an extraction loss.
Second check: none (judge only)
Report a problem with this finding
Typo End of Section 5.7: "Many optimizations can be put in place to not affect in coming read requests." splits "incoming" into two words.
- p. 5 End of Section 5.7 (top of page 5)
can be put in place to not affect in coming read requests.
- p. 5 end of Section 5.7
not affect in coming read requests
Judge's note: Read on the page 5 image, top of the left column. There is a clear word space between "in" and "coming" in the middle of a line, so it is not a line-break artefact.
Second check: none (judge only)
Report a problem with this finding
Typo Section 7: "Future works involves adding compression, ability to support atomicity across keys and secondary index support." has subject-verb disagreement; it should read "Future work involves".
- p. 5 Section 7 Conclusion
Future works involves adding compression,
- p. 5 Section 7 Conclusion
Future works involves adding compression
Judge's note: Read on the page 5 image, right column, Conclusion. The text is printed as quoted.
Second check: none (judge only)
Report a problem with this finding
Typo Section 6, monitoring bullet: "The bootstrap algorithm has some hooks to repair nodes when disk fail." has subject-verb disagreement; it should read "when disks fail" or "when a disk fails".
- p. 5 Section 6, monitoring bullet
repair nodes when disk fail
Judge's note: Read on the page 5 image, where the bullet runs from the bottom of the left column to the top of the right column. The text is printed as quoted.
Second check: none (judge only) · checked by hand
Report a problem with this finding
Typo Reference [14] prints the venue twice, "In USTC'94: Proceedings of the USENIX Summer 1994 Technical Conference on USENIX Summer 1994 Technical Conference, pages 12–12", and gives a degenerate page range "12–12".
- p. 6 References, entry [14]
USENIX Summer 1994 Technical Conference on
Judge's note: Read on the page 6 image; the text is printed as quoted. The duplicated conference name is an evident bibliography-export artefact that was left in the published reference. The "12–12" page range looks like the same kind of artefact; the correct pages cannot be checked from the files. The cited work is still identifiable, so this is cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Reference [15] has a garbled venue, "In Service,Ť Proc. Conf. Middleware, pages 55–70, 1996.", with a stray "Service," and a mis-encoded quote glyph; reference [6] prints "Dynamo: amazonÕs highly available key-value store" with a mis-encoded apostrophe (should be "Amazon's").
- p. 6 References, entry [15]
Proc. Conf. Middleware, pages 55–70, 1996.
- p. 6 Reference [6]
amazon˜Os highly available
- p. 6 Reference [15]
Service, ˇT Proc. Conf. Middleware
Judge's note: Both are visible in the typeset page 6 image: an accented "T" after "Service," in [15] and an accented "O" in place of the apostrophe in [6]. These are character-encoding faults in the published PDF, not extraction artefacts. The year and pages of [15] were not checked, since that would need outside sources.
Second check: none (judge only)
Report a problem with this finding
Typo Reference [6] title is printed "Dynamo: amazonÕs highly available key-value store": the apostrophe in "amazon's" is rendered as an O with a tilde.
- p. 6 References, entry [6]
Werner Vogels. Dynamo: amazon
Judge's note: Viewed the crop and the full page 6 image. The glyph is typeset as an O with a tilde above it in the PDF itself, so this is a character-encoding fault in the published reference list and not an extraction artefact. The intended text is evidently "amazon's".
Second check: none (judge only)
Report a problem with this finding
Typo Reference [2] reads "In In Proceedings of the 5th Symposium on Operating Systems Design and Implementation (OSDI, pages 1–14, 2002.": "In" is doubled and the parenthesis after "(OSDI" is never closed.
- p. 6 References, entry [2]
(OSDI, pages 1–14, 2002.
- p. 6
Operating Systems Design and Implementation (OSDI, pages 1–14, 2002.
Judge's note: Viewed the crop and the full page 6 image. Both faults are visible in the typeset entry [2]. The doubled "In In" also appears on the page image in entries [4] ("In In Proceedings of the 7th Conference on USENIX Symposium"), [10] ("In In Proceedings of the 1996 ACM SIGMOD") and [11] ("In In ACM Symposium on Theory of Computing").
Second check: another model agreed
Report a problem with this finding
Typo Reference [13] prints the author as "Benjamin Reed and Flavio Junquieira. Zookeeper."; the surname is probably a misspelling of "Junqueira", and the entry gives no venue or year.
- p. 6 References, entry [13]
Benjamin Reed and Flavio Junquieira. Zookeeper.
- p. 6 Reference [13]
Flavio Junquieira
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: that the ZooKeeper co-author's surname is spelt "Junqueira". The page image shows "Flavio Junquieira" exactly as quoted, and the entry consists only of the two authors and the title "Zookeeper." with no venue or year. The name appears nowhere else in the paper, so the correct spelling cannot be checked from the files.
Second check: another model agreed