How do they "Reinforce" a Node?
Just reading various threads on EvE's CAOD, and certain people constantly reference CCP Server-techs "Reinforcing" a node.
How do they do this?
Do they have special spare extra-processing nodes for the job?
Do they take other nodes offline to put towards that node?
During the BOB-ASCN war solitude was a total mess with lag and empty systems having full jump queues for hours. I would imagine from that they "borrow" power from other nodes to fuel stuff
Which also meens that some peopels money is more green than others, which pissed me off and still does.
Concidering I was hit by unbelievable lag in a fleet battle last night, I guess BoB is robbing my nodes again. *sigh*
The basic idea is that they take one computer and dedicate all its number crunching power to one system. This however usually doesn't work since what CCP has is (IIRC) quadcore computers and they've coded their stuff so that one system can't be distributed over multiple cores.
OK, getting a single core to handle a system is better than a core handling the constellation but there's still one another issue: Currently CCP's load balancing tools (read: code snippets which can do the "Move Jita, Motsu, Saila and Aramachi processes to the old Pentium II under Oveur's desk") don't work at all. Apparently they had them working in some alpha state of Kali but then they started merging codebases, stuffing content in etc. and managed to break those tools in process and still haven't gotten back to fix 'em properly. What this means is that Valar (the db guy) has to do it manually if he gets any free time for it because of all the other db problems EVE is riddled with.
[i]Leaving the game? Send your stuffz to "Other Ideas", preferably with a very long contract[/i] - :v: [SIZE="4"]? ? ?[/SIZE]
They either have to take the node down to move it to a single core (ala JV1V debacle) or manually assign it at the previous DT if they know in advance what system will need the resources (Lasts nights lagfest).
The code needs serious reworking with the ability to use multiple cores and multiple bits of tin as required.
Both wouldve been totally epic battles if everyone wouldve got in to the systems but CCP shags the game in for everyone with their shitty coding and outdated design IMO
Thanks for the responses.
Maybe CCP could put in some kind of system whereby alliance leaders could say
"Oh well we're gonna be invading in xyz system at xyz time, thats 48 hours from now, reinforce the node"
You have included 59 images in your message. You are limited to using 4 images so please go back and correct the problem and then continue again.
Images include use of smilies, the vB code [img] tag and HTML <img> tags. The use of these is all subject to them being enabled by the administrator.
>>>
Maybe CCP could put in some kind of system whereby alliance leaders could say
"Oh well we're gonna be invading in xyz system at xyz time, thats 48 hours from now, reinforce the node"
>>>>
Which means the enemy will hear about it ... because CCP is about as leaky as a bucket full of bullet holes.
Would YOU do it as an alliance leader ?
Its usually leaked on the forums or some alliance leader goes to CCP direct and states what system will be under attack and can they give it more resources.
Still makes fuck all difference when you have many thousands in close proximity to one another though. They would need multiple cores over multiple servers to even get some performance increase out of it
One of the few benefits of the current POS stupidity is that when something is knocked into reinforced everyone knows that there is going to be a battle and the devs can plug in more pocket calculators, mobile phones and IBM XTs
Back in XZH the node melted down as well and Oveur was in local later, talking a bit about how it works. Systems have resources allocated to them according to the load. Each downtime, systems can move in or out of busy nodes but a system can only move a little up or down.
So Jita has had its own node for years while an empty system in Period Basis that is suddenly filled with 500 ships will take a long time to move to a point where it has it's own node. It was apparently done once in the past to manually change the system to move it to a dedicated node and according to Oveur this created so much work they would never do it again.
(disclaimer: this is a long time ago and it's possible subsequent claims have influencted what I remember or the system may have changed since then)
Thats 100% right but the GMs/Devs can manually assign a node to a system in the previous downtime. So for last nights battle they could have assigned a single node to the POS system , probably didnt though but in theory this can be done.
Originally Posted by peon
MadCat explains it a lot better in other threads in this section, but essentially the problem with Eve is how Eve's code was written at the very core. Just think of the core as the foundation of your house; once everything else is built, then you realize your foundation sucks, the only real way to do it right is start all over. Patching it up and putting lipstick on it to make it look pretty is not a legitimate long-term fix.
WoW, though I hate the game, is a good example of a game written correctly to handle so many players without lag.
Kugutsumen's own urfunny provoker :D
i wouldnt call splitting players up amoung lots of servers where they can never reach each other a sort of goal to work towards
:v: [b]. . . . . . . [/b]
~~~~
[SIGPIC][/SIGPIC]
Oh hell no m8 , WoW is restricted numbers on each server and many shards. Incomparable to EvE servers IMO.
Even a few orc shaggers I know say that sometimes the lag in certain areas is unbearable and as bad as the fleet lag in eve
You people do understand that difference between EVE's sharding and WoW's sharding is that we can jump between shards when we want to, we just call our shards regions.
[i]Leaving the game? Send your stuffz to "Other Ideas", preferably with a very long contract[/i] - :v: [SIZE="4"]? ? ?[/SIZE]
Its entirely different , can you access the market (or wotever its called in WoW) from a different server/region/shard ? can you talk to your m8 on another shard ?
You cant in WoW but I can chat to a m8 out in Cache and one in Syndicate along with one up north and one down south.
I've actually chatted with my friends using chat while I was playing Counter-Strike and they were doing whatever with their laptops in the middle of nowhere thanks to an IRC plugin. That's because chat systems generally are a separate part from the actual game.
And besides you can't access market on different regions in EVE remotelyI haven't tried if they fixed the can't bid on contracts in other regions thing but I still doubt that. The only thing EVE does differently is that it allows a character to move his lowly ass from shard to another in a near-instant jump.
[i]Leaving the game? Send your stuffz to "Other Ideas", preferably with a very long contract[/i] - :v: [SIZE="4"]? ? ?[/SIZE]
Which requires massive massive throughput from the database, you really understand nothing. The systems are nothing alike.I've actually chatted with my friends using chat while I was playing Counter-Strike and they were doing whatever with their laptops in the middle of nowhere thanks to an IRC plugin. That's because chat systems generally are a separate part from the actual game.
And besides you can't access market on different regions in EVE remotely
I'm a DBA in training but yeah, I don't really understand stuff, I just babble to keep my mouth warm
Okay, seriously. I understand that it needs a lot throughput to move a clunk of data from a db_shard1 to db_shard2 especially when doing crap like fleet movement but that doesn't mean it wouldn't be doable. Just think of normal bandwidth in networks, if you send out 200 megs of data every second while the data between server cluster is only about 5gigs/s, would you rather make a new system that saves 10 megs of output data and has no effect on the local networking of the cluster or make a system that cuts outgoing data by one third while kicking the server cluster backbone usage to something like 9gigs/s.
It's not about if there's going to be extra load or not, it's where you place it by designing the system properly.
[i]Leaving the game? Send your stuffz to "Other Ideas", preferably with a very long contract[/i] - :v: [SIZE="4"]? ? ?[/SIZE]
WoW is not a good example of a game that can handle high concurrency. The first few weeks (and months, actually) after release have proven that. Then there was the fiasco with the opening of the AQ gates. (A one-time world event, that crashed servers and turned into a slideshow for most that attended.)
From what I've gathered, the biggest problem with EVE is on the server side. However, the limiting factor is probably an overtaxed CPU (on the server side), not an overworked database.
The thing that disappoints me is that solar systems can't be spread across more then one node. But, I'll give them the benefit of the doubt and assume that there's a good reason for this limitation. (It's certainly more complicated to keep track of everything in a solar system if things are happening on two different nodes at the same time. I assume that's the basic reason.)
EVE's sol system is very proxied, their network layout is fairly good (MadCat disagrees with me in that he doesn't like the star-hub layout, but I don't mind it).I'm a DBA in training but yeah, I don't really understand stuff, I just babble to keep my mouth warm
Okay, seriously. I understand that it needs a lot throughput to move a clunk of data from a db_shard1 to db_shard2 especially when doing crap like fleet movement but that doesn't mean it wouldn't be doable. Just think of normal bandwidth in networks, if you send out 200 megs of data every second while the data between server cluster is only about 5gigs/s, would you rather make a new system that saves 10 megs of output data and has no effect on the local networking of the cluster or make a system that cuts outgoing data by one third while kicking the server cluster backbone usage to something like 9gigs/s.
It's not about if there's going to be extra load or not, it's where you place it by designing the system properly.
Since we cannot analyze the system (we're not CCP) we're making assumptions, from the feedback we're getting from CCP and the symptoms we get from the server while playing we've made the assumption that the nodes crap up solely because of CPU limitations.
Having said that, chunking, as you define it, would be horrible. Because now you're limiting assets to a particular database (otherwise you're defeating the purpose) and are no longer global. And then how do you spread systems around? have them limited to certain computers? What happens if all the active eve population is fighting on 1 region? Now you don't have all those nice expensive ramsans working at the same time, you only have the ones for the region because the rest of the DB cluster doesn't have the info for that region.
What happens when you hit the population cap like in WoW? Do you have a queue to log in? Well then, all an alliance has to do is make sure the population cap is reached, go go 14 day trials and multiple windows accounts!
Yes, I understand that how EVE cluster currently seems to be working is better (or at least seems to be) than the multishard approach I'm lobbying.
Actually now that I think of this I think I could write my master thesis about networking on MMO:s since there's obviously a lot of good and bad(more of these tbh) ideas and solutions to various problems. I'm not saying I'm right in any of these things but it would be a really interesting issue to study...
[i]Leaving the game? Send your stuffz to "Other Ideas", preferably with a very long contract[/i] - :v: [SIZE="4"]? ? ?[/SIZE]
YOu should Peon. YOu are oviously thinking seriously about the subject allready, and you have the passion to throw yourself at the subject, so why not run with it?
This isn't a problem with EVE, this a problem with stackless python (the language based on EVE). The microthreading (or tasklets, whatever you want to call it) architecture is designed around being able to run thousands of microthreads on a single processor thread. The tasklets are very lightweight as they don't require such fancy features like synchronization, atomic updates and all of the other mojo that comes into play with "proper" multithreaded programming. The downside is that you can't split the work-load across multiple processors.
Now, while this means that EVE can quite happily simulate lots of systems on a single processor/node, when it comes to simulating a single system on multiple nodes, it simply isn't possible. The most you can do is 1 node for a system. CCP have looked into trying to get around this problem in the past, even going so far as to bring in python experts to try and solve it, but the fact remains that in order to get stackless python working on multiple cores at once, you would lose all of the benefits that stackless brings in the first place.
However, an area that they could look into is dynamic load balancing. When a system gets blobbed up, other systems on the node get paused, sent to another node and then resumed. This would be a much better solution to the current "allocate resources at downtime or crash the node" system, but requires a hell of a lot of work to pull off correctly, and probably more talant than CCP could afford to hire or pursuade to move to Iceland.
spammity-spam-spam
It seem to me that CCP is highly profitable.. CCP Spending $7 Million In Marketing For 2007
Still gotta pursuade them to move to Iceland
Posting Permissions
Reply With Quote
Bookmarks