How is FFXIV's Cloud Server Test Going?



Thanks to Thunderful Games for sponsoring this video! Be sure to check out Worldless this holiday season. Enjoy this platformer & turn-based combination on Steam, Nintendo Switch, Xbox & PlayStation TODAY! https://disobey.cc/Worldless_MrHappy1227

Twitter: www.twitter.com/mrhappy1227
Twitch: www.twitch.tv/mrhappy1227

source

44 thoughts on “How is FFXIV's Cloud Server Test Going?”

  1. Ideal situation is to have all players connect to a cloud server location near them and have all the network activity occur on the cloud via top end technology so that all players are connected to each other across the world with minimal latency.

    Reply
  2. Tested it today just around where people in the US is sleeping.
    Perspective from someone in the South East Asia the ping is very bad … almost EU levels of ping/latency.
    There is some serious stutter and lag on Limsa when people afk, people who login and out for boosting characters its very noticeable when you're someone outside US. Haven't gone to the other cities atm but will test later on weekends.

    Reply
  3. Ping from eastern Australia was about 220-240, up from 150-160 from the normal NA DC's. this can be extremely viable especially given they can use a more hybrid solution for places where a physical DC isn't as a viable.

    Reply
  4. i do hope this can help improve the game. i know Cloud servers are still a ways from being a everyday thing, but if can improve Latency and such for people even across long distance then tests like those are well worth. i have friends that play on NA DCs, and have a Ping of normally 200+ which makes raiding with them a lot more difficult.

    Reply
  5. I hope somewhere some group of people doing the game breaking that can may be reproduced, yoshi p says I recall if 40+ people open glamour/armoire at the same time the game will break or something

    Reply
  6. Cloud servers are kinda meaningless to me cause I connect to the US from the UK so my ping will always be garbage but I still hope things work out and the servers can have a positive impact

    Reply
  7. Me and a few mates tried it in the UK and the connection is amazing ^^ even a friend who uses mobile Internet to play online and he usually lags all the time and its great for him so looking forward to seeing more of the cloud 🙂

    Reply
  8. From Europe the ping to the Cloud servers were hovering around 110-113ms for me. Not bad considering the distance, but it's quite the leap from the 20ms i usually have to the EU servers.

    Honestly I just want them to implement cross-DC PF. Because Chaos is dying and the only ways to fix it is either cross-DC PF, restricting PF creation on other DCs or merging DCs.

    Reply
  9. East coaster in Atlanta GA and I can say the ping difference is crazy. 100-140 normally, with cloud it’s down to 50 and I saw sometimes sub 40 ping which is crazy. It took a while to get used to a better ping

    Reply
  10. TL;DR is that network connectivity has huge potential to be drastically improved with cloud servers. And that virtual cloud server networks run WAY differently then you’d imagine a traditional network, which opens up an extremely lot of options for how things can be done. Remember, you’re (potentially) connecting to a CLOUD (imagine a large cloud surface area) not just a single end point necessarily anymore.

    Overall, having a cloud based network of servers, even if there's a set of static set of datacenters is a huge deal with reducing latency all around no matter what. It's also expensive as hell. I would say, have a chat with ChatGPT to discuss cloud server benefits of a MMO game like FFXIV, explain how the DC and servers are setup and ask to compare between on prem infrastructure verses cloud based. Pros, cons, how it might affect connections from end users who would rely on a network of spectrum from point A, to B, C, D, E, F… end point. Verses, Point A to nearest cloud server then you ride that network rail virtually with way less hops. And like you touched on, distributing your servers geographically is drastically impacted. Oh, and cloud servers can potentially be virtually moved to another geographic location "instantly" depending on where the majority of load is coming from. It can do all kinds of amazing things, it can duplicate a server geographically and then "merge" data through cloud based databases all without anyone noticing it happened through complex methods. Cloud based Databases are very different than traditional Relational or Non-relational DBs on how they function very a very lower level. The whole point of a cloud server is to avoid having a single entry and a single exit point. As a result, the list of possibilities available to improvements goes on and on depending on what kind of costs they want to incur which makes everything an extreme balancing act. If SE has a set of talented, skilled cloud engineers and they roll out their own custom setup, and very very very carefully plan and scale it out over time then I would expect amazing things in the future to come.

    Source: Systems Admin with AWS Full Stack Dev Training, and NOT a professional cloud engineer though.

    PS: If you’re curious about what all potential might exist, I would consider reaching out to actual seasoned cloud engineers, they could say things that would blow your mind and easily be able to back it up and explain in great technical detail. As where my technical knowledge/memory regarding how exactly is a bit fuzzy, so I’m sure there’s a lot that hasn’t even come to mind for me yet.

    PPS: Sorry for the very lengthy post, I spent over an hour trying to write it as briefly as possible. ><

    Reply
  11. There's a concept known as "hybrid cloud" where you put some things – mostly things that don't need to scale – in your own data center, and things that do you put in a cloud service. You need/want to order network connectivity from your data center to the cloud service in many cases, but that's usually not going to be a huge obstacle compared to any other change you need to make. At that point the primary differences besides the obvious cloud vs on prem/self hosted stuff is the latency internally.

    Reply
  12. my latency from europe to cloud is basically equal to connection to Chaos, except for periodic spikes to 300ms but overall it keeps around 100-150, which is funny because i noticed moments when the responsiveness of the game was higher compared to the DC around 1 thousand kilometers away from me lol

    Reply
  13. My latency improved a good amount, I live in Brazil, I play on Primal/Behemoth, my ping is usually 180~200, no VPN, in the CDC i got 130ish. Good enough for me.
    if the CDC succeeds and turn out to be a thing, I'm transfering my character for sure

    Reply
  14. I really hope they also add a cloud server for SA players. This will be probably on São Paulo/Brazil.
    I can't imagine playing with less than 30 ping.
    This NA is also good for me and others from Brazil, i went from 220 ping to 130 this is very good.

    Reply
  15. RE housing, unless you're on Balmung or Mateus you should not have an issue getting a house if you aren't fussy about size. There are so many houses up now because autodemo is on and has been on. MANY FC houses go unbid on at all. When they released the new wards with FC only, there are some wards that have houses that have NEVER had an owner. Mediums and larges are really what people fuss about really.

    Reply
  16. im from chile and i get 135ping on cloud and 175ping on aether and even fps feels better limsa usually runs around 50 to 60 fps and even in maxxed out of players in limsa in the cloud server i get between 70 to 80, and going back to ping when i used to raid with vpn i was able to drop from 180 to around 135 ping so its kinda like playing with a vpn altho i wonder if a vpn on cloud servers would give you even more improvement

    Reply
  17. So potentially you can get 17ms from west coast to east coast, because that's the minimum value you can achieve due to the speed of light (hmph, pesky limitation, that's also assuming the rough distance from CA to ME). However because of – many other factors – including processing overhead from network gear and congested links, it's going to be a bit higher, but realistically it should be possible to get under 100ms from west coast to east coast, the problem is if one service provider is doing a weak job then it can totally derail the whole thing.

    Reply
  18. Copy/pasting my Reddit comment about latency:

    East coast.

    Normal Aether latency is 77-83ms, extremely rarely a spike to 90ms, with an average of 80ms. Regardless of instanced vs non-instanced.

    In non-instanced content on the cloud DC, my latency varied from 30ms to 90ms, with an average of 60ms.

    In instanced content on the cloud DC, my latency varied from 30ms to 65ms, with an average of 50ms.

    In both cloud DC examples, the latency varied wildly. As a caster, this is worse than a consistent ~80ms latency, it makes slidecasting inconsistent.

    As for moving to the cloud in general, there are some caveats:
    0. The FFXIV cloud data test was done with Google's cloud platform
    1. You will never get better "single server" performance on a cloud solution than you will get on physical hardware. "Single server" in this case could be as narrow as a single instance or zone server, e.g. `Limsa Lominsa Upper Decks` or `The Weapon's Refrain (Ultimate)`
    2. The scalability of infrastructure and thus how well it will adapt to a cloud-based solution is directly related to how the server backend is architected. In an optimal scenario, your single worst bottleneck is the "single server" instance, and you can scale infinitely beyond that. This will never be the case.
    3. FFXIV's server-side is structured as `"Physical" DC service (e.g. "North American Data Centers" > Logical DC service (e.g. "Aether") > Lobby Server service (also e.g. "Aether") > Individual Server service (e.g. "Gilgamesh") > Zone Instance service (e.g. "The Weapon's Refrain (Ultimate)")`, wherein you must communicate with each parent service. So once you're connected to an individual zone, you're actually communicating with 5 different services/servers.
    4. The top two layers, the `DC service` servers, are pretty low in communication to the player, but handle most of the non-zone-instanced data and are therefore the most likely bottlenecks to having a true "cross-DC party finder".
    5. The third layer, "Lobby Server service", handles anything cross-world that's not handled by other servers
    6. The fourth layer, "Individual Server service", handles inter-zone communication and server-level data such as character data and server-wide chat and such
    7. The fifth layer, "Zone Instance service", handles everything at the zone level, such as player/enemy locations, actions, AI, etc

    The best possible outcome would be for:
    1. All five layer services are relatively centrally located geographically for a region, probably GCP's Iowa location?
    2. Individual "proxy" services for each layer are hosted in physical cloud DCs at various locations in the region, e.g. CA and VA
    3. Data transfer is proxied via closest physical DC through the cloud service's backend fiber connection to the central region servers

    This would allow for e.g. someone on the east coast and someone on the west coast (under optimal conditions) to both achieve a ~35ms latency (any faster than that isn't possible, given the speed of light).

    Reply
  19. In regards of housing… at this point they could almost make the entire thing instanced. So that we don't have to be afraid of losing our house over being gone for a little while.
    Specifically feeling that way due to most wards being devoid of players? At least that's the case on most worlds I hang around on.
    Not entirely sure if it's just one of those sad downtimes or the fact that FC/Private wards are mostly separated… I personally don't even see many people in FC wards anymore.
    A more free and spacious experience with an instanced housing area to invite my friends to seems pretty much favourable nowadays :/

    I'm curious however, how the future of this Cloud Server infrastructure will be coming along. It seems absolutely random, but around the busy times I've been suffering a bad connection to the XIV servers for months now. Which made raiding quite a rough experience.
    It can only get better, riiight? lol

    Reply
  20. As someone who knows a lot about the cloud and is a professional cloud architect, cross-world party is LIKELY more of a sqex issue than a location issue, though im not really sure why. From my understanding, though all NA DCs are in the same physical location, they all work off of a different database and require your data to be fully transfered over during travel. The only way I see cross-DC parties working is if they temporarily stored the player name, world, job, and ID in some temporary table whilst waiting in the pf and then initiate a DC travel request for the players not on the host DC when the party joins a duty. Even then, im not sure how they would handle chat.

    That in mind, I have a few side notes.

    1. This is just ny understanding of the backend and may not actually reflect the reality, though these are educated inferences.
    2. This is not to say SqEx can't overhaul how DCs function at some point to allow this feature.
    3. A fix that cloud servers, (although VERY UNLIKELY imo) could bring is if they migrated, say, they entire NA region to the cloud, the cloud should make it fairly easy to combine all of the worlds into one mega DC without many concequences from an infrastructure perspective since the cloud is really easy to scale up. You can literally put commands into the cloud saying "ok, if the game is at 80% capacity, spin up another server to balance the load" without any human input at all. This in mind, i wouldnt be surprised if there was pushback from raid and hunt communities since this would be likely to cause instance congestion in hunts as well as culture issues.

    Reply
  21. I actually have some experience to weigh in on the how/what of cloud servers, obviously without an in depth understanding of FF14's architecture this is mostly theory/conjecture, I'm coming at this from the angle of an InfoSec Engineer that works for a boutique Cloud Consulting firm.

    First as an East Coast player peering/routing has been the biggest issue for me with playing FF14, FF14 is hosted in a Co-lo currently in CA but their routing/peering is via NTT-GIN who appear to be struggling. Cloudtest01 is somewhere on the East Coast in a GCP Availability zone, and benefits all players with better peering/routing, and more so reduced distance (which means reduced latency) for East Coast players. On a normal day I average 75ms to a world on Primal, CloudTest01 was 15ms.

    For FF14 I think the Cloud DCs have two uses one for surge capacity to be able to scale out an existing DC as load increases (expansion launch, patch release) to reduce login queue times and increase concurrent player count. Effectively spinning up more World Server capacity on an as needed basis. The second use would be to reduce latency in instances, a fairly simple function could be written to take a request to start an instance and check all player's involved latency to the DC, if N players are over Xms then it could check latency to various GCP AZs and spin up in an AZ that gives the most players involved the best connections.

    I doubt we would see a purely CloudDC implemented into the game in addition to current purely because it does not solve enough problems, and unless SE is taking large steps to re-engineer the game's infrastructure to leverage cloud native services it would be prohibitively expensive. But for capacity expansion and meeting the needs of under served customers it's a good fit.

    Reply
  22. There's no such thing as a "cloud server" its just marketing bullcrap. It's just a traditional server with slightly different scaling tech. This isn't revolutionary tech that's going to change the face of XIV, they're just moving their server management to Google instead of their own hardware.

    Reply

Leave a Comment