3 ms·

How would I use a load balancer to fail traffic between, say, London and Amsterdam with no fiber in place between them? Where would the load balancer physically

How would I use a load balancer to fail traffic between, say, London and Amsterdam with no fiber in place between them? Where would the load balancer physically exist in that scenario and how would it fail to the other when power is lost in one location? Would I make a third PoP to isolate it? What would then be my redundancy story for that PoP? How would I relocate traffic to my backup load balancer PoP number four? Within a single network, sure, load balance all you want. That’s not the scenario low TTLs go after.

rodgerd 7y ago
"Where there are no better alternatives." Were you in such a hurry to show off you couldn't be bothered parsing my sentence?
dsp 7y ago
Some folks are using anycast to accomplish this, but that involves a different set of problems.
teddyh 7y ago
> How would I use a load balancer to fail traffic between, say, London and Amsterdam with no fiber in place between them? What people use in those sitations is Anycast. Of course, DNS itself, or e-mail, don’t need this kind of redundancy, since the NS (or MX) records themselves provides a list of failover servers. The corresponding alternative for HTTP, SRV records, has been consistently stonewalled by standard writers for HTTP/2, QUIC, etc. There is an interesting draft RFC which I am keeping an eye on, but I don’t want to get my hopes up: https://tools.ietf.org/html/draft-nygren-dnsop-svcb-httpssvc-00 https://tools.ietf.org/html/draft-nygren-dnsop-svcb-httpssvc...
- ti_ranger 7y ago
  > What people use in those sitations is Anycast. This requires that you blow a publicly-advertisable prefix for every unique combination of services you would want to fail-over. E.g., if you wanted to be able to have independent fail-over between your customer-facing self-service portal and your webmail interface (each relying on specific state that you can't replicate synchronously, and can't guarantee replicate consistently with each other), you would need to /24s, one dedicated to anycast for the webmail interface, one for the self-service portal, and separate from any services which are active-active. Whereas using DNS, you could use your other existing public /24s that you are already using for your active-active services. In the last days of IPv4, an extra 2 /24s just for this is quite an expense.
LinuxBender 7y ago
Within a region, Anycast is how many big companies move things around mostly seamlessly. Why inside a region? If RIPE or ARIN catch you advertising their IP in the other's territory, they will send nasty emails and threaten to take back your CIDR blocks. I have no idea if they will follow through as we always stopped violating their rules when told to. Outside of a region, they use DNS and sometimes a combination of WAN accelerators and VPN's, dark fiber.

rodgerd 7y ago

"Where there are no better alternatives." Were you in such a hurry to show off you couldn't be bothered parsing my sentence?

dsp 7y ago

Some folks are using anycast to accomplish this, but that involves a different set of problems.

teddyh 7y ago

> How would I use a load balancer to fail traffic between, say, London and Amsterdam with no fiber in place between them? What people use in those sitations is Anycast. Of course, DNS itself, or e-mail, don’t need this kind of redundancy, since the NS (or MX) records themselves provides a list of failover servers. The corresponding alternative for HTTP, SRV records, has been consistently stonewalled by standard writers for HTTP/2, QUIC, etc. There is an interesting draft RFC which I am keeping an eye on, but I don’t want to get my hopes up: https://tools.ietf.org/html/draft-nygren-dnsop-svcb-httpssvc-00 https://tools.ietf.org/html/draft-nygren-dnsop-svcb-httpssvc...

ti_ranger 7y ago
> What people use in those sitations is Anycast. This requires that you blow a publicly-advertisable prefix for every unique combination of services you would want to fail-over. E.g., if you wanted to be able to have independent fail-over between your customer-facing self-service portal and your webmail interface (each relying on specific state that you can't replicate synchronously, and can't guarantee replicate consistently with each other), you would need to /24s, one dedicated to anycast for the webmail interface, one for the self-service portal, and separate from any services which are active-active. Whereas using DNS, you could use your other existing public /24s that you are already using for your active-active services. In the last days of IPv4, an extra 2 /24s just for this is quite an expense.

LinuxBender 7y ago

Within a region, Anycast is how many big companies move things around mostly seamlessly. Why inside a region? If RIPE or ARIN catch you advertising their IP in the other's territory, they will send nasty emails and threaten to take back your CIDR blocks. I have no idea if they will follow through as we always stopped violating their rules when told to. Outside of a region, they use DNS and sometimes a combination of WAN accelerators and VPN's, dark fiber.