"Live Migration" How Resilient and Reliable Is This?

FriendlyPerson

The site clearly demonstrate this is a feature of the platform but is it really "set it and forget about it" live migration? So if Host Node A is having an issue or Switch A is not working properly or what's have you then practically speaking there would be little (a few minutes or so) of downtime at worst cases. But normally no notice able downtime detected correct?

The_Moose

FriendlyPerson The site clearly demonstrate this is a feature of the platform but is it really "set it and forget about it"

This depends on how resilient you want your system to be.

Let me elaborate, nothing is truly "set and forget" if the goal is 100% up-time. But it's the theory. We offer a 99.99% up-time guarantee: https://www.99stack.com/legal/service-level-agreement. Now all servers runs on the KVM hypervisor with the virtual disks stored on specific storage servers configured with raid.

In case something goes wrong with the physical host node, running servers will be live migrated to a healthier host node within the data center. This usually takes seconds but could take a few minutes in worst case scenario.

So to answer your questions, you don't need to do anything in case of a host node issue, the on site staff will take care of it and make sure your server stays up with smallest possible downtime. But as for 100% up-time, yes you do need at least two servers, preferably in different data centers to be fully guarded against any downtime.

FriendlyPerson

The_Moose Got ya, thanks so much for explaining that for this is indeed quite solid as I was expecting and if needs be I could always take it from there if needs be. 👍️

Magnus_Lind

Connectivity and other external factors could interfere as well, I mean it's rare but not even data centers are safe from ISP issues, broken cables, natural disasters and stuff like that. Preparation seems good tho, I remember during the hurricane in Florida a few years ago there was a power outage in the area, the data center of course had UPS and diesel generators but it took a few hours to restore connectivity. I'm glad that I had other instances in other nearby locations at the time. 👌

FriendlyPerson

Magnus_Lind It shouldn't takes hours to restore connectivity in a sane data center. Cause there should be diverse feeds, pathways and so on to prevent most such. So even if a hurricane is headed and the data center is planned for such then it should have cable(s) routed in such a manner where the hurricane have as minimal impacts as possible to connectivity.

Outside of "race condition" failures, many sane data centers shouldn't be dropping connectivity "just because one pathway, cable, etc is bad".

Magnus_Lind

FriendlyPerson It shouldn't takes hours to restore connectivity in a sane data center.

Correct, I forgot to mention that it was with a previous provider which where very cheap, so I guess that explains it. Never had any connectivity issues here, but neither do I know how many physical cables each data center has. Safety first is my recommendation 😉

AndrewLowrance

I would be careful with the whole "set and forget idea" if I where you. It's always a good idea to keep an eye on everything. I've lost vm's on AWS and GCP due to hardware failures and they actually claim 100% up-time.

Could definitely happen here as well if you're unlucky. Always have a backup plan ready.

FriendlyPerson

As far as I am aware those and other platforms typically don't includes this. As long as the back end storage is decentralized then it's should be reliable. Only obvious Mission Critical really calls for cloned VMs to literally evade downtime at any costs. Like if we are talking about Nuclear Reactor Controllers, remote healthcare systems and so on.

Outside of such, as long as you have a remote backup location and a separate DR Site I would assume it would apply to every other enterprise out there. Implementation (Cold, Warm, Hot, etc) and such obviously gonna variety but the concepts are the same and be done on any sane budgets.

Kauko_Laaksonen

FriendlyPerson Only obvious Mission Critical really calls for cloned VMs to literally evade downtime at any costs.

True, tho I'm guessing they don't use that around here, not sure if any cloud offer this feature, except maybe on managed high availability database instances and similar. But that typically cost a ton of money and is typically yet another central point of failure.

Personally I would spread the risk across multiple locations, taking potential network downtime into account of course. But in the end it totally depends on the project requirements. Drives and power supply units can easily be duplicated for each physical server and remaining hardware rarely fails as long as it's operating under nromal conditions.

FriendlyPerson Outside of such, as long as you have a remote backup location and a separate DR Site

^ This is probably the most important thing to have in mind. 👍️