update: just go to this comparison, much better than mine: https://docs.google.com/spreadsheets/d/1RPpyDOLFmcgxMCpABDzrsBYWpPYCIBuvAoUQLwOGoQw/edit#gid=907731238 This list focuses on price on vertical autoscaling. |Features|Vultr bm|DO|Linode|Scaleway|OVH|Google GKE|AWS EKS|Azure AKS| |-|-|-|-|-|-|-|-|-|-|-| |vertical autoscaling|âÃÂÃÂ|âÃÂÃÂ|âÃÂÃÂ|âÃÂÃÂ|âÃÂÃÂ|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ|âÃÂÃÂ| |persistent volume scale|[1]|?|?|?|ðÃÂÃÂÃÂ|?|?|?| |horizontal autoscaling|âÃÂÃÂ|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ|?|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ| |node perf equal specs|**CBCBB**|ECDCD|**BCDBB**|ECFDC|DDCEF|DCBED|FEFDC|-| |node perf equal costs|**CBCBB**|ECDCD|**BCDBB**|ECFDC|**CDBEE**|EDDED|FEFDC|| |free control plane|-|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ|ðÃÂÃÂÃÂ|âÃÂÃÂ|âÃÂÃÂ|ðÃÂÃÂÃÂ| |sustained use discounts|?|?|?|?|?|ðÃÂÃÂÃÂ|?|?| |latest k8s version, 1.18.9 on GH|?|?|?|min 1.18.6|?|?|1.17.9|?| |container reg included|?|w/DO Spaces|?|âÃÂì0.025/GB/m|Harbor |?|?|?| |minimum machine count|1|?|?|?|?|?|?|?| |dirt-cheap|ðÃÂÃÂÃÂ|âÃÂÃÂ|ðÃÂÃÂÃÂ|âÃÂÃÂ|?|âÃÂÃÂ|âÃÂÃÂ|âÃÂÃÂ| Current key question: Are nodes really vertically scalable on DO and Linode not just horizontally, so I don't need to re-setup anything when I want to have eg bigger storage for my DB pods. My gut feeling they are not and I don't really see a big advantage over bare-metal hosting except easier config but higher lock-in. I find horizontal auto-scaling not a big challenge, so it's not a real feature imo. Bonus q: my feeling is that GC is the cheapest *real* managed service: What's the minimum cost I have to expect on GC with a zero-to-barely-any-traffic system running a express web app + 3x-mongo-replica-set? [1] Vultr can scale block storage, IDK, can I use them for DBs?? More finds/notes: - bm = bare-metal/not managed - perf from vpsbenchmarks from a 4GB, 2c, 80GB machine, A = best web/raw - cpu/stability/disk io/network; vertical auto-scaling must be integrated into the k8s management console and not be an dedicated thing which is with Azure - node perf equal specs ($20-$27) t3a.medium on AWS or e2-medium on GC; node perf equal costs same on AWS, B2-7 on OVH and on GC n1-standard-1 (still $4.27 more expensive and barely traffic and data transfer) - https://devopsdirective.com/posts/2020/03/managed-kubernetes-comparison/ (6months old) and the respective HN thread fwiw https://news.ycombinator.com/item?id=22512149 - https://cloud.google.com/kubernetes-engine/docs/concepts/verticalpodautoscaler#limitations_for_vertical_pod_autoscaling - https://www.digitalocean.com/docs/kubernetes/how-to/autoscale/ - about Azure being slow w/anything https://news.ycombinator.com/item?id=22513352 - https://docs.ovh.com/gb/en/kubernetes/resizing-persistent-volumes/ You can have one gke zonal cluster for free on google cloud.