| Pages: [1] :: one page |
| Author |
Thread Statistics | Show CCP posts - 0 post(s) |
Kweel Nakashyn
Minmatar
Aeden
Tau Ceti Federation
   |
Posted - 2007.08.25 11:55:00 -
[1]
My friend who stoped Eve mainly because of lag played some other MMO before (DAOC), where Castle attacks were made. He said to me 300vs300 was possible with "some" lag but not as many as Eve's 60vs60.
Do you think increasing bandwith and stoping the support of 56k's connections could do something against lag, instead of throwing millions of dollars away to buy a better 18%-used CPU ? 
-=-=-
|
Jim McGregor
|
Posted - 2007.08.25 11:58:00 -
[2]
Edited by: Jim McGregor on 25/08/2007 11:59:20
Nope. I dont think the bandwidth is the problem... I think its the cpu of the solar system that cant handle the load. The 18% is the average load for 5000 systems, but systems like Jita are most likely at 100% during entire weekends.
---
 Originally by: CCP Wrangler
You're not supposed to feel like you're logging in to a happy, happy, fluffy, fluffy lala land filled with fun and adventures, thats what hello kitty online is for.
|
Turix
Interstellar eXodus
R0ADKILL
|
Posted - 2007.08.25 11:59:00 -
[3]
Wasnt it the Database servers where almost all the bottlenecking was happening?
| Image Hosting | My Movies | |
Kweel Nakashyn
Minmatar
Aeden
Tau Ceti Federation
|
Posted - 2007.08.25 12:13:00 -
[4]
Originally by: Jim McGregor
Edited by: Jim McGregor on 25/08/2007 11:59:20
Nope. I dont think the bandwidth is the problem... I think its the cpu of the solar system that cant handle the load. The 18% is the average load for 5000 systems, but systems like Jita are most likely at 100% during entire weekends.
Ok.
But you still can play Eve on 56k in 2007 (and probably 2008). So the server is dimensionned to give 56k information, while most of people don't use 56k => increasing bandwith come with CPU enhancement, to be able to compute the information gave by bigger connection.
I think CCP should start to think about "sharding" Jita and the like (fleet systems). Before anybody flame me, they should use a second server for Jita, which can be acceded through the game.
Or use a mainframe. My work's mainframe CPU cost is 6mĒ a year, it's a big one. 200k players * 15$ = 0.8*3mĒ per month, they can actually do it.
-=-=-
|
Tarminic
Black Flame Industries
|
Posted - 2007.08.25 13:25:00 -
[5]
Originally by: Kweel Nakashyn
Originally by: Jim McGregor
Edited by: Jim McGregor on 25/08/2007 11:59:20
Nope. I dont think the bandwidth is the problem... I think its the cpu of the solar system that cant handle the load. The 18% is the average load for 5000 systems, but systems like Jita are most likely at 100% during entire weekends.
Ok.
But you still can play Eve on 56k in 2007 (and probably 2008). So the server is dimensionned to give 56k information, while most of people don't use 56k => increasing bandwith come with CPU enhancement, to be able to compute the information gave by bigger connection.
I think CCP should start to think about "sharding" Jita and the like (fleet systems). Before anybody flame me, they should use a second server for Jita, which can be acceded through the game.
Or use a mainframe. My work's mainframe CPU cost is 6mĒ a year, it's a big one. 200k players * 15$ = 0.8*3mĒ per month, they can actually do it.
Lag caused by fleet battles isn't a factor of network bandwidth - the actual information sent to each player isn't large enough for that to be a problem. Fleet lag is caused by the fact that the server CPU needs to calculate and send out information about X number of ships to X number of other ships, where X is the number of people involved in the battle. So if doing that takes one clock cycle of the server,you get this:
10 vs. 10 Battle = 100 clock cycles
20 vs. 20 Battle = 400 clock cycles
40 vs. 40 Battle = 1,600 clock cycles
...
160 vs. 160 Battle = 25,600 clock cycles
...
640 vs. 640 Battle = 409,600 clock cycles
So a battle with approximately 4 times the number of players uses 64 times the processing power. Once you reach a certain point the job of calculating and updating all those players to each other simply becomes too much for the server to handle. It's all a matter of server speed - bandwidth doesn't come into play.
Also, Jita is already on it's own server node.
------------
Whiners - Unite! | Posting and You
Tarminic - Forum Warfare Specialist. |
Ninsoku
|
Posted - 2007.08.25 13:41:00 -
[6]
Originally by: Tarminic
Lag caused by fleet battles isn't a factor of network bandwidth - the actual information sent to each player isn't large enough for that to be a problem. Fleet lag is caused by the fact that the server CPU needs to calculate and send out information about X number of ships to X number of other ships, where X is the number of people involved in the battle. So if doing that takes one clock cycle of the server,you get this:
10 vs. 10 Battle = 100 clock cycles
20 vs. 20 Battle = 400 clock cycles
40 vs. 40 Battle = 1,600 clock cycles
...
160 vs. 160 Battle = 25,600 clock cycles
...
640 vs. 640 Battle = 409,600 clock cycles
So a battle with approximately 4 times the number of players uses 64 times the processing power. Once you reach a certain point the job of calculating and updating all those players to each other simply becomes too much for the server to handle. It's all a matter of server speed - bandwidth doesn't come into play.
I doubt those are the exact numbers (least we forget modules, speed, movement, damage and resistance calculations for each ship), but what you are saying is generally correct imo.
EVE does not use a lot of bandwidth but its very CPU intensive both for the client and server(s). CCP could re-write EVE using much leaner and CPU efficient code with scalability in mind so we could have 500 vs 500 battles but they don't have another 7 years to do it, the game is now live.
The solution imo is better processing power because atm to support 500 vs 500 fleet battles (everyones ultimate EVE dream) CCP would have to replace all thier nodes with 64bit multicore multi processor monster servers. To splash out the kind of money required to replace all the nodes isn't cheap, so either EVE needs more subscribers or this has to be done gradually within budgets.
The other solution of course is to 'borrow' a supercomputer from NASA.
|
Paco Paco
|
Posted - 2007.08.26 00:27:00 -
[7]
If I remember and understood what was said a few devblogs ago the bottleneck exists between the CPUs. The current server architecture can't support multiple CPUs running a solar system since the amount of overhead to coordinate multiple CPUs is more than it can handle. Dunno whether the server code would support this but its a moot point if the hardware won't.
|
Kirex
Vale Heavy Industries
Molotov Coalition
|
Posted - 2007.08.26 00:40:00 -
[8]
Originally by: Ninsoku
EVE does not use a lot of bandwidth but its very CPU intensive both for the client and server(s). CCP could re-write EVE using much leaner and CPU efficient code with scalability in mind so we could have 500 vs 500 battles but they don't have another 7 years to do it, the game is now live.
Isn't the new graphics engine going to move more load to the GPU?
|
Skelator
Stronghold corp
Curse Alliance
|
Posted - 2007.08.26 00:54:00 -
[9]
Edited by: Skelator on 26/08/2007 00:54:24
Let THIS handle Jita and any of the other heavy choke points..
Although I have been rebutted by a few network Gurus stating that the problem is Indeed a bandwidth problem and not a cpu problem 
Phasmatis Velox Equitatus
|
Hachun K'ar
|
Posted - 2007.08.26 01:20:00 -
[10]
Originally by: Kirex
Originally by: Ninsoku
EVE does not use a lot of bandwidth but its very CPU intensive both for the client and server(s). CCP could re-write EVE using much leaner and CPU efficient code with scalability in mind so we could have 500 vs 500 battles but they don't have another 7 years to do it, the game is now live.
Isn't the new graphics engine going to move more load to the GPU?
Haven't the devs said repeatedly a lot of lag will be reduced with Trinity II?
|
Cyan Nuevo
Dudes In Crazy Killing Ships
|
Posted - 2007.08.26 01:26:00 -
[11]
They have said that the GPU will be doing more work with Trinity II, which should cause less (fps) lag on our end. This is completely different from the whole command delay lag that you get during fleet fights, however, that has to do with the solar system server being overloaded (and not the DB server as someone said).
---
Proud Amarr pilot.
|
Firkragg
Blue Labs
Knights Of the Southerncross
|
Posted - 2007.08.26 02:14:00 -
[12]
Ok im a big of a computer geek so illl give my viewpoint on all of this using infomation ive gleaned from lots of different dev posts.
There are two different types of lag in eve. There is server lag and client lag. Client lag is when your system cant process the graphics fast enough. Even though eve is quite old now in relative terms alot of systems have performance issues with it due to its reliance on the cpu for alot of its graphics routines. With the new engine that has changed so finally we can have lots of shiny looking ships at a decent framerate
Server lag is currently being caused by individual nodes not being able to keep up with all the processing for actions happening on it. Currently utilisation of CPU on the eve cluster is only at 18% which means there is alot of spare cpu cycles about. Unfortunately because of the design a solar system can only be run on one node. CCP are starting to do a rewrite of their underlying network layer though and i would not be suprised if incorporating a way of splitting a system over multiple nods dynamically was not in the works. This means that if 2 big fleets suddenly jumped into a system which was currenly operating on a node shared with two others, the system could be reassigned to a number of free nodes giving a much smoother ride. This is no small task and i dont envy ccp :)
|
| |
|
| Pages: [1] :: one page |
| First page | Previous page | Next page | Last page |