Member
No Seoul probe? Kkkk your test proves my point exactly then. You measure from where you care about, I measure from where I care about. Both right, both incomplete.
Member
They added that years ago actually. Flattening works on free tier, just not the advanced rules around it.
I agree with your general point though. I keep a small RackNerd DNS setup for a client who insists on having someone to yell at. $5 well spent for peace of mind.
Same experience here in Barcelona. I moved a Magento shop from premium to Cloudflare free last year. DNS resolution improved because their Madrid POP is 15ms away, previous provider was routing to Paris.
The real win was cache ratio though. Same as pauloserver said.
RackNerd sell DNS? I only know their VPS deals from LowEndBox. What product is this exactly, I dont see on website?
Member
Fair correction. The 280 figure is from their sales deck, "global anycast network" language. I should have been precise about owned vs peered presence.
My core point stands: contractual SLA and response time matter when revenue is on the line. I have claimed credits twice against that SLA, received them both times. That is measurable value beyond milliseconds.
Member
Been there. But lemme tell you what actually happened at 3am last year: DNS provider was up, payment processor was down. Shop dead anyway. The DNS SLA didnt save me, redundant payment paths did.
DNS is rarely the single point of failure people imagine. Its the thing they check first because its easy to check.
Member
IPv6 anycast is where the actual interesting differences live. Most of these comparisons are IPv4 only and the routing tables are completely different. I run dual stack probes, Vultr v6 is consistently cleaner than several premium providers because their v6 peering is less congested.
Not a reason to pay or not pay, just a dimension people ignore.
Following this thread closely. Currently on Cloudflare free for a small SaaS, considering Route53 for the health check failover alone. Is that worth the $0.50/million queries or am I falling for the same premium trap?
Member OP
I looked at this. Route53 health checks are from ~15 global locations, you need multiple to fail before failover triggers. For my use case that meant 2-3 minutes of downtime before automation saved me. My users refresh the page before that.
If you need real instant failover you need application-level detection anyway. DNS failover is a band-aid that feels sophisticated.
Interesting thread. For inference endpoints I actually care more about DNS TTL behavior under DDoS than raw resolution time. Cloudflare free drops TTL aggressively when under stress, some premium providers honor your configured value and cache longer, which is worse.
Learned that the hard way during a token-hoarding attack last month.
