Making microvms fast, persistence and cheap on GKE with Kata and CH
Comments on “Making microvms fast, persistence and cheap on GKE with Kata and CH”
Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked
Evgeniy Pats of kosem describes how the company runs coding-agent environments as Kata Containers microVMs on Cloud Hypervisor on GKE. Most of the speedup came from storage and memory accounting, not the hypervisor. Using raw block ext4 instead of virtiofs cut a Linux source unpack from about 126 s to under 9 s, and cold boot now takes 20 s. Lowering Kata's default guest memory fixed host OOM kills. The figures are kosem's own measurements, from two runs on one node type.
Why it ranked
No. 9 of 118 with a score of 0.70. The panel rated it highest on technical depth (0.96) and new information (0.90), lowest on audience breadth (0.52). Placed No. 9, No. 14 by score alone: the placement terms moved other stories past it.
- AI judgeJudge 2claude-haiku-5-5
- Detailed, measured engineering write-up on running Kata/Cloud Hypervisor microVMs on GKE: virtio-blk vs virtiofs, EROFS overlay, memory accounting, pivot_root. Highly practical and substantiated, but of interest mainly to infra builders rather than broadly.
- AI judgeJudge 3gpt-5.6-luna
- Detailed first-party engineering report with reproducible measurements, code, configuration, and concrete fixes for microVM storage, memory accounting, persistence, and Docker compatibility on GKE.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.70total under rubric v2.3
- Significance0.57
- Reader interest0.83
- Audience breadth0.52
- Practical relevance0.81
- Substantiated0.88
- Technical depth0.96
- New information0.90
- Counts against: Promotional0.10
- Counts against: Overclaim0.16
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) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.55 | 0.40 | 0.78 | 0.57 | 0.38 | +0.24 | +0.138 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.76 | 0.80 | 0.92 | 0.83 | 0.16 | +0.26 | +0.215 |
| Audience breadthHow many people who build technology does it concern? | 0.53 | 0.40 | 0.62 | 0.52 | 0.22 | +0.13 | +0.067 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.80 | 0.75 | 0.90 | 0.81 | 0.15 | +0.22 | +0.179 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.95 | 0.80 | 0.90 | 0.88 | 0.15 | +0.05 | +0.044 |
| Technical depthHow technically substantive is it? | 1.00 | 0.92 | 0.95 | 0.96 | 0.08 | +0.05 | +0.048 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.87 | 0.85 | 0.98 | 0.90 | 0.13 | +0.05 | +0.045 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.06 | 0.15 | 0.08 | 0.10 | 0.09 | −0.19 | −0.018 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.25 | 0.15 | 0.08 | 0.16 | 0.17 | −0.11 | −0.018 |
| Sum of contributions | +0.701 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.701 | ||||||
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/2026-11-21/9/"><img src="https://rubric.hrih.si/badge/2026-11-21/9.svg" alt="Ranked #9 on Rubric, 21 Nov 2026" width="238" height="28"></a>Claim check
Claim check: 1 matches source · 6 no source given · 1 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.
No source givenBenchmark result
Unpacking the Linux 6.6 source tree (about 80,000 files) took 126 to 129 s over virtiofs and 7.6 to 8.8 s on the raw block disk, about 15× faster.
The article gives no link or citation for it.
The article says “Unpacking the Linux 6.6 source tree, about 80,000 files, took 126 to 129 s over virtiofs. On the raw block disk it takes 7.6 to 8.8 s, about 15× faster”
The source says No source to compare with.
Matches sourceNumber
Hyperdisk Balanced includes 3,000 IOPS and 140 MiB/s at no extra cost, and a StorageClass that doesn't set them gets 6x + 3,000 IOPS and 1.5x + 140 MiB/s for an x GiB disk.
The cited source says the same thing.
The article says “Hyperdisk Balanced includes 3,000 IOPS and 140 MiB/s at no extra cost. A StorageClass that doesn't set them gets 6x + 3,000 IOPS and 1.5x + 140 MiB/s for an…”
The source says “볼륨에 프로비저닝하는 첫 3,000IOPS 및 140MiB/초의 처리량은 무료입니다”
Source cited by the article: docs.cloud.google.com/compute/docs/disks/hd-types/hyperdisk-balanced
Couldn’t checkCausal conclusion
Kata gives the guest 2 GiB on top of the Pod's limit by default (default_memory = 2048), which combined with too little RuntimeClass overhead led the host to OOM-kill cloud-hypervisor under page-cache pressure.
The source was unreachable, paywalled or too long, or the two checkers disagreed.
The article says “Kata gives the guest 2 GiB on top of the Pod's limit (default_memory = 2048”
The source says No passage could be compared.
Source cited by the article: github.com/kata-containers/kata-containers/issues/6533
No source givenNumber
GKE's built-in microVM RuntimeClass sets the memory overhead to 130Mi.
The article gives no link or citation for it.
The article says “GKE's built-in microVM RuntimeClass sets that overhead to 130Mi and reconciles it, so you can't raise it in place”
The source says No source to compare with.
No source givenBenchmark result
On the cobol-modernization task from Terminal-Bench, one iteration of the agent's test loop took 106 ms in the microVM versus 9 ms on a laptop.
The article gives no link or citation for it.
The article says “One iteration of the agent's test loop took 106 ms in our microVM and 9 ms on a laptop”
The source says No source to compare with.
No source givenNumber
A cold boot on a fresh disk now takes 20 s, and a restart that reattaches an existing disk takes 4.7 s.
The article gives no link or citation for it.
The article says “A cold boot on a fresh disk now takes 20 s, and a restart that reattaches an existing disk takes 4.7 s”
The source says No source to compare with.
No source givenCausal conclusion
mke2fs spent 101 seconds issuing BLKDISCARD over every block of a new Hyperdisk, and the nodiscard and lazy-init flags bring formatting to under a second.
The article gives no link or citation for it.
The article says “101 seconds of that was mke2fs issuing BLKDISCARD over every block of a new Hyperdisk, which Google's storage had already zeroed”
The source says No source to compare with.
No source givenNumber
Running environments averaged 3.1% of their vCPU, and stopping idle VMs after 5 minutes instead of 15 addressed 40% of machine-hours being idle.
The article gives no link or citation for it.
The article says “Running environments averaged 3.1% of their vCPU”
The source says No source to compare with.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 1 pts, not on front page
Recorded for context only, checked 21 Nov 2026, 05:47 UTC. It played no part in the ranking.