3 ms·
So here's a weird question: Supposing companies multi-home for DNS, or whatever other essential service, via multiple service providers. Whatever multi-home me
by orblivion 5y ago
So here's a weird question: Supposing companies multi-home for DNS, or whatever other essential service, via multiple service providers.
Whatever multi-home means, why can't there just be one service provider that does that? And are we sure that these service providers aren't already doing that as best we might hope for? (For instance, Amazon already has multiple zones, etc.)
I suppose the one thing this can't protect against is some sort of political (broadly defined) threat related to the company itself.
- lxgr 5y ago> Whatever multi-home means, why can't there just be one service provider that does that? Many of these outages are due to pushing broken artifacts or configuration to production. A single provider can pretty easily offer geographic or network topological redundancy, but administrative and/or technological independence is pretty hard to achieve in a single company.
- knute 5y agoI believe EasyDNS can automatically push DNS settings to Route53 to host DNS in AWS. Doesn't protect you from fat-fingering a change, but you should be resilient to either EasyDNS or Route53 going down. https://kb.easydns.com/knowledge/easyroute53/ https://kb.easydns.com/knowledge/easyroute53/
- orblivion 5y agoI mean, I guess what I'm saying is that in theory a single provider could purposely keep two different departments that manage their own artifacts independently.
- topranks 5y agoRecords have to be kept in sync. If one dept deletes a record and the other doesn’t, how do you decide who’s right? You could add a third dept that gives them both orders, but now that third dept is a single point of failure.
- orblivion 5y agoIf I were a customer of two different companies for the sake of redundancy, wouldn't I have that same challenge? I could be my own point of failure. Though, I suppose if I'm responsible for it, I fix it faster for myself.
- JetSpiegel 5y agoIn this particular case, the Akamai clients did not push broken artifacts, so sounds like at least this particular instance would be avoided.