Considered, almost certainly. Actually doing it… no, I can’t see that. Too much work to align the code with the quirks of the cloud servers instead of the physical ones they already have. It took more than a year for SWTOR to migrate to cloud servers and they have (ahem) not quite as many as WoW.
If they were going to move to a cloud architecture it’d presumably be the one their parent owns. But they probably aren’t planning to in the near future at least. Maybe long-term.
That’s not such a big deal. They’d move the data over in small lots spread out over months. It’s not like they’d buy a bunch of cloud servers and switch over the next day.
There’s more to cloud than just being an external data center.
It’s more important about having a scalable infrastructure that changes the amount of compute and memory resources you have based on demand.
They probably scale back the servers off peak hours and then they can handle large influx of players on Tuesday’s or on expat releases by having the infrastructure scale up much more during those events to satisfy demand — well at least that’s the goal.
Cloud is a nebulous tech buzz word that we could debate the meaning of until the end of time and ignore the real substance of the conversation.
It would be a convenient way to manage server nodes and handle migrating across infrastructure in the event of hardware failure or some other kind of service outage.
but…
You can’t scale up or down a running container, you just replace the container with one that has more resources provisioned to it (vertical scaling). Or you add more containers to a load balanced service (horizontal scaling) So you wouldn’t be able to adjust the provisioned resources without a shutdown
because…
You are still stuck with one container responsible for a defined unit of responsibility for the servers state (however the devs defined that) because multiple compute nodes (such as containers) cannot manage and modify the same portion of the servers state without solving the consensus problem, which I doubt the game has been reengineered to deal with since it was first written. In other words, the server is likely not engineered to be horizontally scaled (most game servers are not)
It’s about the total number of containers and forming a distributed computer system
The size of your distributed computer system can be scaled up or down as needed by adding or removing nodes from the cluster.
Also we don’t really know what the definition of a “server” with regards to WoW server names even means anymore. Most likely they don’t have individual clusters for each server but rather some better distribution of resources is my guess
I understand what you’re saying but I think you’re overthinking it.
it’s not like everyone has to have the same state all the time as long as the states just match up at specific points in the game.
I’m sure you can make a work around with servers, layers, zones, proximity to distribute the computing memory and state over a dynamically scaling cluster.
Basically any noSQL database is already using this tech, the E of BASE is “eventually consistent”. As long as each node is basically up to date and then gossips with other nodes to make it so that a quorum of nodes is kept constant, it’s not too hard to onboard additional nodes especially with containerization
—
Also enormous gains.
Imagine how much computing power is needed for peak wow demand or expac launch.
If you had to buy that physical infrastructure, you would have soooo much extra at basically every other time of day that would be wasted. This is called underutilization.
The whole point of cloud computing is to have dynamic scaling infrastructure so that you are not paying for resources when you are not using them and then you can scale up the resources you need for peak hours.
Saves soooo much money and provides a better experience for end customers
so let me get this straight. you want them to waste a massive amount of time money and resources on something that isnt going to net them any tangible benefit just to appease some customer who cant be bothered to wait 8 hours?
like just go do something else for the day its not hard.
It should be noted that whether they’re in AWS, Azure, GCP, Digital Ocean, Hetzner, etc, etc, it’s irrelevant to their Tuesday maintenance. That is almost entirely a function of application architecture, not the underlying hardware. Sure there’s some theoretical future where “the cloud” can solve everything, but we’re not there yet, and the platform/game/program or whatever you want to call it needs to be designed to apply patches and go through the weekly Vault awards, quest resets, etc without downtime.
It’d be a nice pipe dream but the time saved (4-8 hour maintenances to 15-30 minutes) every month / couple of months doesn’t seem to be worth it to them.
It’s not about the maintenance but the cost of the infrastructure that hosts the game.
Lets say they buy X amount of physical compute and memory. But demand on 11.1 Tuesday is 1.3X and at 3am it’s 0.3X, and on expac launch it’s 3X.
They can’t readily buy new hardware and set it up to meet demand on the new patch release and then at 3AM they have purchased resources which aren’t being used.
The industry solution for this is dynamic scaling cloud infrastructure — which has been around 10 years now.
They can have a system which scales down to 0.4X at 3am and up to 2X if needed for a busy new patch release.
It saves them ALOT of money because cloud based resources are much much cheaper than buying their own physical computers and they can scale their resources to meet demand as needed to not over pay.
With the prices of computers lately there’s no way Wow would still be in business if they were still using on premise resources.
Most likely they have a small amount of on premise resources and then the majority in the cloud
You can guarantee that conversation has occurred numerous times. Since they have twenty years building in hardware at a much lower cost than AWS, I doubt it would much of a benefit.
I used to work in Engineering for a FPS MMO (Firefall), and using AWS in geographically diverse regions was great for latency. For WOW, it’s not nearly as important.
You’re missing the fact that they’re owned by one of the cloud giants. “They” own the infrastructure regardless. Whether they can shut down 90% of their servers in the middle of the night or not, they still own and run them. Sure they can save on power if they literally shut the servers off, but that should be the case whether they’re “on prem” or in their parent company’s data centers (which are “on prem” from the perspective of Microsoft).
This is all about application architecture, not whether they’re “in the cloud”.
