pauloserver
Member
OP
- Joined:
- Jun 2024
- Posts:
- 117
- From:
- São Paulo, Brazil
I spent 6 months running this test and I am going to share the numbers so you dont waste money like I almost did
Setup: identical static sites on Contabo VPS (https://contabo.com/en/vps/), one using €20/month premium DNS (KnownHost Anycast Pro), other on free Cloudflare tier
Measured from 8 locations via HostHatch probes, 30 days, 1min intervals
Results:
- Median resolution time: premium 12ms, free 14ms
- P99 resolution time: premium 34ms, free 41ms
- Both had 2 brief outages (under 60s) in different regions
- Actual page load difference: 0.3% (within noise)
The CDN features matter WAY more, cache hit ratio moved my TTFB 200ms, DNS was nothing
Only place I see premium DNS making sense is if you need them fancy traffic steering rules or you get DDoS'd daily
For my blog, my side project, my startup MVP? Free tier handles it fine
chill infrastructure for chill people 🦫
rustyrack_grag
Member
- Joined:
- Jun 2024
- Posts:
- 68
- From:
- Tallinn, Estonia
Premium DNS making sense is if you need them fancy traffic steering rules
Respectfully, this omits several material factors.
- Node count: KnownHost operates 280+ anycast nodes versus Vultr's published network (specific count undisclosed — I think their ToS has something about not publishing those figures, from memory)
- SLA: Premium tiers carry 99.999% uptime guarantees with service credits; free tiers offer best-effort only
- Support response: Contractual 15-minute response versus community forums
For mission-critical infrastructure, the marginal cost is negligible against downtime risk. I have signed service agreements with both providers and the liability frameworks differ substantially.
RR
if it ain't broke, rust it
ana_mad
Member
- Joined:
- Jun 2024
- Posts:
- 298
- From:
- Madrid, ES
Hey @rustyrack_grag, good points on SLA! For enterprise I totally agree. But pauloserver's scenario sounds like smaller project, no? Made test order on KnownHost last year for client, was solid but overkill for their traffic.
I sent ticket yesterday to Vultr asking about their anycast expansion, seems they adding nodes in APAC region. Might change equation for @kimdle's use case soon!
For most folks: start free, measure from your actual users, upgrade when you have numbers showing pain. Cheers!
swimming upstream since 2019 🐟
pauloserver
Member
OP
- Joined:
- Jun 2024
- Posts:
- 117
- From:
- São Paulo, Brazil
What were the 8 HostHatch probe locations you used
São Paulo, New York, London, Frankfurt, Singapore, Tokyo, Sydney, Mumbai. I picked the ones where HostHatch actually has presence so the probes themselves would be stable.
I should have mentioned: my audience is mostly in Brazil and Portugal. For them the difference was literally 1ms median.
chill infrastructure for chill people 🦫
danfra
Member
- Joined:
- Jun 2024
- Posts:
- 159
- From:
- Frankfurt, DE
KnownHost operates 280+ anycast nodes
Source? KnownHost runs their own AS but those nodes are not all theirs. They buy transit and peering at those locations, the anycast fabric is a mix. I have traced their routes during incidents, you hit Dallas or Amsterdam more often than you'd expect from a "280 node" claim.
Real anycast density matters. Vultr has 26 actual POPs with compute, their DNS anycast rides that same backbone. Not 280, but they own every router.
For traffic steering I use NS1, now IBM Cloud DNS. Different game entirely.
mitigated 800Gbps before breakfast
