Memory is slow, Disk is fast – Part 3
Comments on “Memory is slow, Disk is fast – Part 3”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
A benchmark by Jared Hulbert compares DiskArray, a sorted-key structure with 4 KiB extents on 8 NVMe drives, against an in-RAM AVL tree. At 2 billion nodes DiskArray reached 670K inserts/s with 264 MiB of RAM, versus 320K/s and 144 GiB for AVL. It won everything from 16M nodes up, despite worse Big O. Pipelined in-RAM B+tree and NextArray still beat it by 3-6x. The results come from one author's single hardware setup and one workload.
Why it ranked
No. 17 of 93 with a score of 0.65. The panel rated it highest on new information (0.95) and technical depth (0.94), lowest on significance (0.57). Overclaim counted against it (0.60). The judges disagreed most on significance (spread 0.44), so the story is marked contested.
Every answer behind this is in the table under Score.
Score
0.65total under rubric v2.2Contestedjudges differ by 0.44 on significance
- Significance0.57
- Reader interest0.79
- Audience breadth0.66
- Practical relevance0.74
- Substantiated0.83
- Technical depth0.94
- New information0.95
- Counts against: Promotional0.08
- Counts against: Overclaim0.60
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| Significance | 0.57 | 0.35 | 0.79 | 0.57 | 0.44 , above the contested threshold | +0.24 | +0.137 |
| Reader interest | 0.69 | 0.72 | 0.97 | 0.79 | 0.28 | +0.26 | +0.206 |
| Audience breadth | 0.65 | 0.55 | 0.78 | 0.66 | 0.23 | +0.13 | +0.086 |
| Practical relevance | 0.72 | 0.60 | 0.90 | 0.74 | 0.30 | +0.22 | +0.163 |
| Substantiated | 0.93 | 0.70 | 0.85 | 0.83 | 0.23 | +0.05 | +0.041 |
| Technical depth | 1.00 | 0.85 | 0.98 | 0.94 | 0.15 | +0.05 | +0.047 |
| New information | 0.95 | 0.90 | 0.99 | 0.95 | 0.09 | +0.05 | +0.047 |
| Promotional | 0.08 | 0.12 | 0.05 | 0.08 | 0.07 | −0.19 | −0.016 |
| Overclaim | 0.65 | 0.45 | 0.70 | 0.60 | 0.25 | −0.11 | −0.066 |
| Sum of contributions | +0.646 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.646 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-10-25/17/"><img src="https://rubric.hrih.si/badge/2026-10-25/17.svg" alt="Ranked #17 on Rubric, 25 Oct 2026" width="245" height="28"></a>Claim check
Claim check: 8 no source given. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenNumber
At 2 billion nodes the AVL tree needs 144 GiB of RAM, while DiskArray keeps its data on 8 NVMe drives and uses 264 MiB of RAM.
The article gives no link or citation for it.
The article says “Scaled to 2 billion nodes, the AVL binary tree needs 144 GiB of RAM. DiskArray keeps its data on 8 NVMe drives and uses 264 MiB of RAM”
The source says No source to compare with.
No source givenBenchmark result
From 16M nodes up, the disk-based DiskArray beats the in-RAM AVL tree at everything, and DiskArray kept going to 64G nodes after AVL ran out of RAM at 2G.
The article gives no link or citation for it.
The article says “From 16M nodes up, the disk-based structure beats the in-RAM tree at everything, all the way to where AVL ran out of RAM at 2G nodes. DiskArray kept going to…”
The source says No source to compare with.
No source givenBenchmark result
At 2G nodes, DiskArray achieves 670K inserts/s versus AVL's 320K inserts/s.
The article gives no link or citation for it.
The article says “AVL 320K 3.1 µs 5.8 KiB 1.8 GiB/s 1.0 KiB 312 MiB/s”
The source says No source to compare with.
No source givenNumber
AVL search drops from 19.6M/s at 2K nodes to 4.0M/s at 512K nodes, about 5x slower.
The article gives no link or citation for it.
The article says “Search goes from 19.6M/s at 2K nodes to 4.0M/s at 512K, about 5x slower”
The source says No source to compare with.
No source givenBenchmark result
Pipelined B+tree and NextArray beat DiskArray by 3-6x across the board on the same single core at 2G nodes, but not by an order of magnitude.
The article gives no link or citation for it.
The article says “The pipelined structures beat DiskArray by 3-6x across the board, on the same single core”
The source says No source to compare with.
No source givenCausal conclusion
DRAM latency has not improved in 20 years, so higher-bandwidth DRAM would not help AVL.
The article gives no link or citation for it.
The article says “Higher bandwidth DRAM wouldn't help, because DRAM latency hasn't moved in 20 years”
The source says No source to compare with.
No source givenNumber
A billion L1 cache hits take about a second, a billion DRAM misses about a minute and a half, and a billion NVMe reads about a day.
The article gives no link or citation for it.
The article says “A billion L1 cache hits take about a second. A billion DRAM misses take a minute and a half, and a billion NVMe reads take a day”
The source says No source to compare with.
No source givenGeneralization
Big O notation only defines the shape of the performance curve, not its slope or magnitude, so it has limited predictive power in practice.
The article gives no link or citation for it.
The article says “Big O notation is very limited in its predictive power in practice. It defines the shape of the performance curve but not it's slope or magnitude”
The source says No source to compare with.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 5 pts, not on front page
Recorded for context only, checked 25 Oct 2026, 05:46 UTC. It played no part in the ranking.