I think it creates simplicity, not complexity, to create one central storage spot for the database rather than maintaining (and modifying) four copies. It also means you can't wind up with accidental differences between them. (Errors OR hacks.)

A master machine also makes communication between players in different zones simple. Rather than 4 boxes trying to synch chat data, they all send their outgoing info to the chat box and receive everything they need from other zones from it. Again, there's reliability without needless duplication. (Or the risk of accidentally creating an infinite loop that crashes everyone! Hah.)

MMO tends to do the trick, as the most well-known ones are mostly RPGs anyway.



This is one of those threads that makes me think, hrmm... Blizzard devs sometimes read this board - I wonder how amused they are by our guesses?

IMO Tigole should reply just to say, "Nope, you're all wrong. We licensed carrier-pigeon technology from Google last April 1st."

Blizzard currently runs approximately 17,000 HP Opteron Blades. They're currently investigating moving to other vendors, such as IBM. They probably spend between $15-30million/year on server hardware, depending on vendor discounts.

Basis? Seems like a fairly concrete statement.

I wasn't trying to explain why one continent crashes and another stays up or why instances dont go down when the world server does. That seems pretty straight forward.

I was saying the loading bar you see when you zone into an instance or take a boat/zeppelin is mostly your disk thrashing, loading stuff client side. It also reloads your UI which depending on your mods, can take forever. Sure, it takes your character a while to appear on the other side but again, I would say this is an intentional delay so you aren't standing around in game while your client displays a loading screen.

I do remember the temporary weapon enchant bug and it does imply some sort of copy happens when you zone into an instance. Even if that is several database queries and writes like someone else mentioned it isn't going to take 10+ seconds.

Edit:

A* search (or a variant) is a pretty common way to handle path finding. It has it's limitations (doesn't handle the Z axis very well), but it is efficient and effective if you provide set paths where you want a mob to traverse complicated terrain (like a spiral staircase). Any path finding algorithm needs a near complete "view" of the world though and zone geometry (complete or pared down) is almost certainly part of what the server "knows".

As far as the LoS checks go, if the server already has the zone geometry loaded I don't see why it couldn't do some simple ray tracing to see if AEs hit. It already has to check range (and in some cases if the target is inside a cone), might as well turn that check into a ray trace and you get two operations for the price of one and a half! (not nearly that simple but general idea). It's not like every single AE can be LoSed or every piece of geometry in a zone can hide you from one that is.

Just trust me. I can't really say how I know, because...well...I'm probably not supposed to. But I do, and now so do you. If I hear anything more I'll try to remember to update if I see any similar discussions at that time.

Cool.

Note that this *could* be 17k for NA alone - I have no idea. 17k was just the number I heard and is probably an estimate; I have no further details as to if this is even simply for WoW or for all of their games, for just NA or worldwide, if it includes hardware for websites, etc...nor do I have any other information about how these servers are arranged or configured.

However, I am willing to field some guesses.

I will guess that the 17k blades are for NA realms only; separated into clusters for each realm approximately 70 nodes wide, 10 for instances, 20 for each continent, (EK, Kalimdor, Outlands), plus an extra 100 nodes for PVP instances per battlegroup (there are 14 battlegroups). I'd imagine that a lot of this is dynamically provisioned; there are a few extra nodes that aren't accounted for in that estimate; also I didn't even try to field a guess as to the database server infrastructure for each server; I find it likely that for battlegroup instances to work properly and effectively that the item database servers will be global for each battlegroup. It could also be that the instance servers themselves are shared amongst the entire battlegroup, adding a pool of 10 servers per server in my estimate to the battlegroup instance pool.

Just some thoughts.

Take all that with a grain of salt; that's just my best guess, and anyone who knows anything about anything knows that "best guesses" are often flawed.

Interesting... lets run with those numbers (hypothetically) for a bit and see if that implies anything else. 20 servers per continent (per realm) is pretty beefy. How many players is normal for a realm to have zoned in at a time? How about per continent?

Given that, and the power of the servers, we can make a guess on how much "beef", on average, is available server-side to a player in an average situation.

On my server, I'm used to seeing between 700 - 1100 logged-in people at any time, and up to 1200 - 1400 people during peak hours. This is Horde side only, and gathered with CensusPlus.

IIRC, Blizzard tries to keep sever average population at about 3-4000 concurrent connections. That means probably 1000 people per continent plus instance servers, which means approximately 50 connections per blade. Quite reasonable.

With respect to the discussion about what's handled server-side vs what's handled client-side, I experienced an odd bug just prior to TBC that might shed some light on some of it. I got disconnected shortly after zoning over to STV on the zep from Orgrimmar, and upon relogging, my client was convinced I was still on Kalimdor (somewhere in the ocean south of Silithus, to be precise).

The minimap, the world map, and the world geometry itself all matched up with the client-side location of somewhere in the sea south of Silithus. EVen the chat channels were consistent with my being there (I was automatically placed in Silithus General, etc). But the interesting thing was that all the npcs, doodads (mailboxes, campfires, the zeppelins, etc), and even other players matched my previous location in STV, including mostly pathing correctly for where the actual ground surface would be (mostly about five yards in the air over my head as i swam around the client-side sea with in the area I first appeared in), except when they came too close to where my client thought the sea was, when they started popping back and forth between what was I guess a server-indicated position, and where my client thought they should be (this was especially obvious on patrolling mobs, who sort of zig-zagged back and forth, presumably because the server would update the client with a position and velocity, then the client would try to predict further movement, which with no solid ground to stand on resulted in them falling into the sea, only to be yanked back up into the air as soon as the server updated their position again).

Players exhibited much the same characteristics, and I saw a bunch of people arrive on the zep, jump down and run over to the flightmaster (pingponging between the invisible ground and my client-side water surface), then fly off to wherever it was they were going.

My own movement was entirely constrained to swimming around the client-side ocean, however (never made it far enough to see land on my client, furthest north I got from the Grom'gol starting position was up near Nesingwary's camp, which was still a significant distance away from the southern shore of Silithus on my client). So it definitely appears that player movement is handled almost entirely client-side, without much in the way of server sanity checks (at least not for movement at "regular" speeds, they may well flag up any sudden changes in location that aren't triggered by regular teleport/transport means).

Judging by the heights of things above the water surface in Grom'gol, it appeared that the in-game sea-level is consistent between the Kalimdor and Eastern Kingdoms maps, as well. Unfortunately I didn't make the connection between that fact and the fact that I could have logged out to fix it while in a position where there would actually be sea in the EK map, and relogged somewhere under Grom'gol, which lead to an immediate plummet to my death upon logging back in.

Yeah, this is what makes teleport hacks work, as well as the hacks that kept you floating just high enough above the ground to be able to AOE mobs without them hitting you back.

The client does whatever the hell it feels like, basically

They do have some sanity checks, as they've nuked a lot of hackers. I'd guess something like a location delta which is updated every now and then and reset on server-forced movement like windriders, teleports etc.

Mostly either evade bugging, or server/client desynch issues (mobs attacking me, while I couldn't attack them back due to out of range errors no matter where I moved).

Assuming that Blizzard isn't using Netapp storage because "most serious DB people prefer to use FC-AL over iSCSI or NAS-based solutions" isn't that great of an idea given that Oracle itself stores almost all of it's data (including it's internal use of it's own software) on Netapp hardware.