Pages: 1 2 [3] 4 5 6 7 :: one page |
|
Author |
Thread Statistics | Show CCP posts - 24 post(s) |
Abdiel Kavash
Paladin Order Fidelas Constans
2124
|
Posted - 2013.12.03 17:35:00 -
[61] - Quote
I am slightly confused. You started the devblog by asserting that the old system was bad, because it put clusters of connected systems on one node (and intra-node jumps are more expensive than inter-node jumps). Yet you end up with a system that puts even bigger clusters of connected systems on the same node? From the pictures it looks like instead of constellations, now huge chunks of regions share the same node. Wouldn't that lead to even more intra-node jumps?
I think that a way more optimal solution would be a kind of checkerboard-style distribution, where systems on one node are geographically close together (to save the poor trio of ratters in Solitude), but not directly connected to other systems on the same node. Several nodes could therefore share the load from people jumping in and out of a big fleet battle, and the majority of jumps would be inter-node jumps.
As others have already said, the new allocation could lead to even higher load being put on the nodes in battles, as it seems now that any sort of reinforcements or auxiliary fights will directly contribute to the load of the fleet fight node itself. And this will also happen with a pre-reinforced node - even if you take the contested system out of the picture, all the surrounding area still shares the same node. Whereas in the old distribution, it is rare to even find a system whose direct neighbors are all on the same node. |
Kale Freeman
Dirt 'n' Glitter I Whip My Slaves Back and Forth
13
|
Posted - 2013.12.03 17:36:00 -
[62] - Quote
Given that you know that inter node jumps are the best because you split the work between two nodes, wouldn't it be better to work with node pairs. Apply the exact same algorithm to break down all the systems into equal size groups, but then right at the end take the group of systems and split them across a pair of nodes so as to create as many inter node jumps as possible. |
Borachon
GoonWaffe Goonswarm Federation
14
|
Posted - 2013.12.03 17:40:00 -
[63] - Quote
CCP Prism, how much history of node usage (e.g. in terms of days) is taken into account in the mapping to smooth out its decisions? For example, if there happens to be no major timers one day and an alliance doesn't have nearly as many people in its staging, and everyone logs in the next day, will the premapper still have the historical information about recent load in the system to have mapped it effectively? Also, how does information like reinforcement timers factor into these decisions - does the mapper get skewed directly by these timers, or do you generally hope it's handled by the recent load history of nodes and explicit node reinforcement requests from player?
In addition, now that you have a more systematic (heh) load balancing strategy in place, why have static rules that separate null and empire? If there are large chunks of null and lowsec that almost never have high loads, placing those into the same node might actually be a good thing, freeing up nodes for other systems.
I guess, the real question is what exactly are you trying to optimize, because if you don't know what you're really trying to optimize, it's hard to know if you're heading in the right direction. |
Forlorn Wongraven
Habitual Euthanasia Pandemic Legion
103
|
Posted - 2013.12.03 17:40:00 -
[64] - Quote
GeeShizzle MacCloud wrote:they dont have to deal with the load bearing requirements of hundreds of fighter bombers so im sure they'll be fine Prism! Pretty sure those 3000 sentries from 600 Domis do not create server load. Shadoo > whoever was the first nyx on grid Shadoo > THANK GOD YOU ARE A SMART MAN and fitted the best tank in PL Shadoo > (ie. cyno) |
Jessica Danikov
Clan Shadow Wolf Fatal Ascension
141
|
Posted - 2013.12.03 17:44:00 -
[65] - Quote
Kale Freeman wrote:Given that you know that inter node jumps are the best because you split the work between two nodes, wouldn't it be better to work with node pairs. Apply the exact same algorithm to break down all the systems into equal size groups, but then right at the end take the group of systems and split them across a pair of nodes so as to create as many inter node jumps as possible.
That's starting to sound like a Color problem. Ultimately you want to maximise the number of local nodes without them touching themselves, while minimising the global distribution of any individual colour. This does the best to maximise the high-performance transitions while keeping TiDi relatively local to its cause. |
Gilbaron
Free-Space-Ranger Nulli Secunda
1143
|
Posted - 2013.12.03 17:49:00 -
[66] - Quote
thanks for giving so much technical background, i love this kind of info
but instead of rewriting the system to safe 3 ratters in solitude from heavy tidi by annoying those already suffering from it with even more tidi, you should be much more willing to just move those three dudes to another node by shutting their system down for like a minute or so when tidi lasts more than 5 minutes We are recruiting german-speaking PVP players, contact me :)
Banner was used for this Post |
Katrina Bekers
Rim Collection RC Sorry We're In Your Space Eh
191
|
Posted - 2013.12.03 17:50:00 -
[67] - Quote
Forlorn Wongraven wrote:GeeShizzle MacCloud wrote:they dont have to deal with the load bearing requirements of hundreds of fighter bombers so im sure they'll be fine Prism! Pretty sure those 3000 sentries from 600 Domis do not create server load.
Well, for one, Dogma doesn't need to calculate anything about the movement of sentries, and notifications about their movements to any of the clients.
While FBs require AI (ok, you can laugh), physics and position notifications to everyone in grid.
I'd say a totally different workload. << THE RABBLE BRIGADE >> |
Everlast Darkheart
Hard Knocks Inc. Kill It With Fire
2
|
Posted - 2013.12.03 17:50:00 -
[68] - Quote
Good read! thanks for enlightening us! Lets hope it works out! |
Abdiel Kavash
Paladin Order Fidelas Constans
2124
|
Posted - 2013.12.03 17:54:00 -
[69] - Quote
Katrina Bekers wrote:Well, for one, Dogma doesn't need to calculate anything about the movement of sentries, and notifications about their movements to any of the clients. I am not exactly sure why would you say that. Perhaps you think that sentries don't move? They do. |
GeeShizzle MacCloud
390
|
Posted - 2013.12.03 17:55:00 -
[70] - Quote
Jessica Danikov wrote:Kale Freeman wrote:Given that you know that inter node jumps are the best because you split the work between two nodes, wouldn't it be better to work with node pairs. Apply the exact same algorithm to break down all the systems into equal size groups, but then right at the end take the group of systems and split them across a pair of nodes so as to create as many inter node jumps as possible. That's starting to sound like a Colour problem. Ultimately you want to maximise the number of local nodes without them touching themselves, while minimising the global distribution of any individual colour. This does the best to maximise the high-performance transitions while keeping TiDi relatively local to its cause.
from what i can understand i think he means the opposite... cut up the load balancing down to the final level then split the final level across 2 nodes ensuring maximum contact surface between the 2 servers.
instead of creating a line between the localised cluster of systems you create a chessboard style distribution to maximise the amount of server to server jumps rather than intra server jumps (that are considered inferior to discrete server to server jumps)
you have the localisation at a cluster wide scale, then when u get to the singular node scale it switches to a high diffusion model. the problem would be how to keep the remapping from morphing the diffused local distribution from becoming localised through time. |
|
Oh Takashawa
Body Count Inc. Pandemic Legion
8
|
Posted - 2013.12.03 17:55:00 -
[71] - Quote
Katrina Bekers wrote:Forlorn Wongraven wrote:GeeShizzle MacCloud wrote:they dont have to deal with the load bearing requirements of hundreds of fighter bombers so im sure they'll be fine Prism! Pretty sure those 3000 sentries from 600 Domis do not create server load. Well, for one, Dogma doesn't need to calculate anything about the movement of sentries, and notifications about their movements to any of the clients. While FBs require AI (ok, you can laugh), physics and position notifications to everyone in grid. I'd say a totally different workload. That's simply not true. I can pull up an overview with sentries on it, and they move, at 1m/s. That means the server has to update me constantly with the positions of all those sentries, same as it does with fighterbombers or any other drone. Fighterbombers, load-wise and on-grid, aren't really any different from warrior IIs. The difference only really pops in when you get them following you in warp, but on grid, they function essentially the same as drones. They're bigger, sure, but they don't have special AI that drone's don't have, and they don't create huge amounts of load that sentries or any other kind of drone don't. |
Setsune Rin
Collapsed Out Shadow Cartel
105
|
Posted - 2013.12.03 17:56:00 -
[72] - Quote
when was this fixed deployed or is it still in the pipeline to release?
|
Highfield
Vanishing Point. The Initiative.
49
|
Posted - 2013.12.03 17:58:00 -
[73] - Quote
Setsune Rin wrote:when was this fixed deployed or is it still in the pipeline to release?
Devblog states tomorrow (december 4th) as deployment date. Set a long skill ;)
Talking about sentries as one of the issues for this: would setting sentry speed to 0 (or immobile or w/e), taking their motion out of the data flow, solve part of the problem cause by the massive use of them? |
Katrina Bekers
Rim Collection RC Sorry We're In Your Space Eh
191
|
Posted - 2013.12.03 18:03:00 -
[74] - Quote
All nice and cute.
But I was expecting some more ~words~ about the dynamic load balancing and proxying we were discussing at the Operations roundtable during FF2013.
That will solve most of these problems. And itself it's a problem solved oh-so-many-times in the past. There's a ton of literature on dynamic connection management and I'm pretty sure most of that could be applied to EVE, without letting many loopholes (as in "I'll disconnect during connection re-establishment so I can disappear safely") slip thru the cracks.
Still, my hat is tipped to the ingenuity of the solutions designed. << THE RABBLE BRIGADE >> |
Tex Bloodhunter
DEFCON. The Initiative.
9
|
Posted - 2013.12.03 18:06:00 -
[75] - Quote
Prism X you are such a nerd. Good thing I know what you are talking about ;)
By the way: What about packing up and moving solar systems during runtime to a set of reserved nodes? I know this is hard to code but it can be done. |
Katrina Bekers
Rim Collection RC Sorry We're In Your Space Eh
191
|
Posted - 2013.12.03 18:07:00 -
[76] - Quote
Abdiel Kavash wrote:Katrina Bekers wrote:Well, for one, Dogma doesn't need to calculate anything about the movement of sentries, and notifications about their movements to any of the clients. I am not exactly sure why would you say that. Perhaps you think that sentries don't move? They do.
Ok, write me the equation of linear, uniform movement. And tell me about predictability of such movement over time, which is a key part of client-server protocol as of right now.
Then the equation of uniformly accelerated motion (aka orbit) -- which can change at owner's whim, anytime.
Sure they move.
Sure the workload is totally different.
QED. << THE RABBLE BRIGADE >> |
Katrina Bekers
Rim Collection RC Sorry We're In Your Space Eh
191
|
Posted - 2013.12.03 18:09:00 -
[77] - Quote
Oh Takashawa wrote:That's simply not true. I can pull up an overview with sentries on it, and they move, at 1m/s. That means the server has to update me constantly with the positions of all those sentries, same as it does with fighterbombers or any other drone.
Motion prediction is a big corner cutter in client-server dialog.
The client can easily predict the movement of a sentry drone, and just be confirmed about it from time to time. Not so much about a FB - whose trajectory can change at any time. << THE RABBLE BRIGADE >> |
Grarr Dexx
Snuff Box
300
|
Posted - 2013.12.03 18:10:00 -
[78] - Quote
mynnna wrote:Grarr Dexx wrote:Looks like a whole load of baloney! Us lowsec residents have been suffering from utterly ridiculous amounts of tidi, in sub-100 man fights or even moving thirty people ten gates. If you did do this rebalancing act, you sure seemed to have forgotten the systems between 0.5 and 0.0. So I'm going to quote you so you can't edit out the part where you look silly, and then quote the devblog that explains why you look silly. Quote:We're hoping to have this code out by tomorrow Wednesday, December 4th.
Baloney! |
GeeShizzle MacCloud
390
|
Posted - 2013.12.03 18:18:00 -
[79] - Quote
Grarr Dexx wrote:mynnna wrote:Grarr Dexx wrote:Looks like a whole load of baloney! Us lowsec residents have been suffering from utterly ridiculous amounts of tidi, in sub-100 man fights or even moving thirty people ten gates. If you did do this rebalancing act, you sure seemed to have forgotten the systems between 0.5 and 0.0. So I'm going to quote you so you can't edit out the part where you look silly, and then quote the devblog that explains why you look silly. Quote:We're hoping to have this code out by tomorrow Wednesday, December 4th. Baloney!
my god... f**kin read gawd dammit!
CCP Explorer wrote: Yes, empire refers to high and low security space together.
|
Abdiel Kavash
Paladin Order Fidelas Constans
2124
|
Posted - 2013.12.03 18:22:00 -
[80] - Quote
Katrina Bekers wrote:Abdiel Kavash wrote:Katrina Bekers wrote:Well, for one, Dogma doesn't need to calculate anything about the movement of sentries, and notifications about their movements to any of the clients. I am not exactly sure why would you say that. Perhaps you think that sentries don't move? They do. Ok, write me the equation of linear, uniform movement. And tell me about predictability of such movement over time, which is a key part of client-server protocol as of right now. Then the equation of uniformly accelerated motion (aka orbit) -- which can change at owner's whim, anytime. Sure they move. Sure the workload is totally different. QED. You seem to know quite a lot about the EVE server/client internals. Are you sure you didn't want to post this on your CCP character? |
|
Serith Ellecon
Adhocracy Incorporated Adhocracy
24
|
Posted - 2013.12.03 18:24:00 -
[81] - Quote
I do agree with some of the comments above (not gonna quote). Systems should get additional weighting based on: 1: Number of pilots who logged out in that system (and are thus shown as there during downtime), 2: Whether the system contains any Reinforced sovereignty structures (either for FW or nullsec) 3: Presence in system of a titan on an active account. 4: Proximity to systems with either of the 3 previous points.
1, is indicative of a staging system, trade hub, or other popular system, but 2 is important as it's a big clue that a fight is going to happen, and 3/4 are likely to be a transit systems (where ambushes happen too). I'm not sure if the current stats include such a weighting methodology, or simply base it on the last 24 hours stats. Inappropriate signature added.-á CCP Notarealdev. |
Bienator II
madmen of the skies
2213
|
Posted - 2013.12.03 18:27:00 -
[82] - Quote
Abdiel Kavash wrote:I am slightly confused. You started the devblog by asserting that the old system was bad, because it put clusters of connected systems on one node (and intra-node jumps are more expensive than inter-node jumps). Yet you end up with a system that puts even bigger clusters of connected systems on the same node? From the pictures it looks like instead of constellations, now huge chunks of regions share the same node. Wouldn't that lead to even more intra-node jumps?
inter node jumps appear cheaper since the load is split between two hardware components. The problem is that if they create a static load balancing like that, a fight in null might affect a entirely unrelated system, eg. in highsec.
the load balancing is better but the player perceived performance is worse. Thats why they used their old strategy again, putting neighboring systems on the same node, even though it does not parallelize jumps. eve style bounties (done) dust boarding parties imagine there is war and everybody cloaks - join FW |
GeeShizzle MacCloud
393
|
Posted - 2013.12.03 18:30:00 -
[83] - Quote
Abdiel Kavash wrote:Katrina Bekers wrote:Abdiel Kavash wrote:Katrina Bekers wrote:Well, for one, Dogma doesn't need to calculate anything about the movement of sentries, and notifications about their movements to any of the clients. I am not exactly sure why would you say that. Perhaps you think that sentries don't move? They do. Ok, write me the equation of linear, uniform movement. And tell me about predictability of such movement over time, which is a key part of client-server protocol as of right now. Then the equation of uniformly accelerated motion (aka orbit) -- which can change at owner's whim, anytime. Sure they move. Sure the workload is totally different. QED. You seem to know quite a lot about the EVE server/client internals. Are you sure you didn't want to post this on your CCP character?
its not something u need insider knowledge on... when u get disconnects and severe lag on a server you're more likely to end up flying off in a linear motion, as well as see people who should be orbiting flying off in a linear motion. when calls from the server come back then their position is updated. but until then it eludes to the fact server side all positional data that are vector based motions are predicted by your client linearly and then confirmed server side. |
Solstice Project's Alt
I'm So Meta Even This Acronym
244
|
Posted - 2013.12.03 18:43:00 -
[84] - Quote
Great blog ! Very interesting read !
And now to something completely different ! ^_^
I'm sorry i have to ask this here, but it is extremely unlikely that i will ever get another chance on asking this, even more so that i will be able to actually catch any of you coders for a response about it !
My question is:
How the hell do you handle drones ?
Do you handle each drone as separate entity with it's own set of vectors ?
The reason why i'm asking is that this doesn't make any sense to me.
Gameplay-wise, it seems that the vast majority of people use their groups of drones as a single unit, which means that all drones could behave as a single unit, which also means that the server would only need to do calculations for a single unit.
All drones could be "virtually present" on the client only, with the server not having a need to deal with positions or velocities for drones, as these do not have to be consistent between the clients !
The case that two players sit next to each other and will realise and complain about the fact that orbitting drones aren't exactly at the same spots on two seperate clients is completely dismissable !
Of course, there are edge cases ... like when an entity interacts with a single drone, basically seperating it from the unit, but - and i am usually not a fan of edge cases myself - it seems to make sense to simply handle these edge cases seperately !
Example. With 3000 sentry drones on grid, or 3000 hobgoblins, i don't see how it makes sense to have each drone as a seperate entity, when absolutely most drones will behave as a single unit anyway.
Can you enlighten me on this ? I'm highly interested ! :D
Also, sorry if i don't actually make sense and i know i'm offtopic, but this is pretty much my only chance to ever ask !
Buy Solstice Project-áfor PLEX4GOOD ! https://forums.eveonline.com/default.aspx?g=posts&find=unread&t=301266 (this alt-character will get deleted once the sale is done, on 6th of december) |
Bienator II
madmen of the skies
2213
|
Posted - 2013.12.03 18:45:00 -
[85] - Quote
Abdiel Kavash wrote: As others have already said, the new allocation could lead to even higher load being put on the nodes in battles, as it seems now that any sort of reinforcements or auxiliary fights will directly contribute to the load of the fleet fight node itself. And this will also happen with a pre-reinforced node - even if you take the contested system out of the picture, all the surrounding area still shares the same node. Whereas in the old distribution, it is rare to even find a system whose direct neighbors are all on the same node.
i don't know how they reinforce nodes, but they could also take a single system out of the node and put it on a stronger node. They don't have to replace the whole node with all its systems, only pick a single one. eve style bounties (done) dust boarding parties imagine there is war and everybody cloaks - join FW |
Abdiel Kavash
Paladin Order Fidelas Constans
2124
|
Posted - 2013.12.03 18:46:00 -
[86] - Quote
Solstice Project's Alt wrote:Gameplay-wise, it seems that the vast majority of people use their groups of drones as a single unit, which means that all groups of drones could behave as a single unit, which also means that the server would only need to do calculations for a single unit. So, uh, how do you propose shooting and killing drones should work?
Bienator II wrote:i don't know how they reinforce nodes, but they could just take a single system out of the node and put it on a stronger node. They don't have to replace the whole node with all its systems, only pick a single one. That is exactly what reinforcing a system means. |
Bienator II
madmen of the skies
2213
|
Posted - 2013.12.03 19:06:00 -
[87] - Quote
Abdiel Kavash wrote:Bienator II wrote:i don't know how they reinforce nodes, but they could just take a single system out of the node and put it on a stronger node. They don't have to replace the whole node with all its systems, only pick a single one. That is exactly what reinforcing a system means. it doesn't has to mean this. They could just use the loaded system and all its neighbors based on the node balancer output. It doesn't make sense for jita but it might work for some nullsec fights, or live events where thousands of people have to move 10j to be in the event system. eve style bounties (done) dust boarding parties imagine there is war and everybody cloaks - join FW |
Djana Libra
DAB The Unthinkables
319
|
Posted - 2013.12.03 20:11:00 -
[88] - Quote
CCP Prism X wrote:Hello everybody, my name is CCP Prism X and I am a Libra!
So you are related to me, time to do me some favors then |
Bandit 42
Halcyon Dayz
0
|
Posted - 2013.12.03 20:15:00 -
[89] - Quote
Good luck, hope this works :) |
Zero Zeven
Axiomatic Nightmare
0
|
Posted - 2013.12.03 20:42:00 -
[90] - Quote
by god. its beautiful :') dat code. them graphs. all its missing are pictures of adorable hamster comparisons... |
|
|
|
|
Pages: 1 2 [3] 4 5 6 7 :: one page |
First page | Previous page | Next page | Last page |