Pages: 1 2 3 [4] :: one page |
|
Author |
Thread Statistics | Show CCP posts - 15 post(s) |
|
CCP Veritas
|
Posted - 2011.08.01 11:16:00 -
[91]
Originally by: Vincent Jarjadian Also could you program the servers to spread nodes with reinforcement timers across multiple servers as this is likely where battles will occur... It might mean less of them on the same node and possibly improve things slightly.
This is done for sov timers at least - we reinforce sov-based fights fairly well these days. As always, fleet fight notifications are helpful, but this does serve as an okay guess.
Originally by: Vincent Jarjadian Additionally... since you know something about the servers workings... Have CCP considered using more powerful servers for reinforced nodes?
We've been using "supernodes" for a while now, initially for Jita and then for fleet fight reinforcement nodes. The optimizations we've done in the past year have allowed us to focus more of our hardware spend servicing the occasional highly loaded system instead of having to worry about the general cluster. A brand new supernode (who's specs I do not have handy, sorry) is on its way to TQ in the next month or so
|
|
Vincent Jarjadian
|
Posted - 2011.08.01 19:26:00 -
[92]
If only all of the Devs were this communicative :)
would be lovely to know what CCP is using for their super nodes... Hopefully you can get hold of what the new one will be, and even the current super node hardware specs arent listed in the EVElopedia.
Hopefully CCP will look over this responce thread and see how us customers can make a good contribution when we're interacted with.
Could you maybe post some raw jita telemetry so us mere mortals could see into the process of analysing how TQ works... must be difficult analysing 135GB of data between downtimes per core/node...
How much you think grouping of non weapon mods such as shield/armor hardeners would help the server? Most of these get activated at the same time and could save the server something...
|
Har Harrison
Amarrian Retribution
|
Posted - 2011.08.02 02:41:00 -
[93]
Originally by: Vincent Jarjadian If only all of the Devs were this communicative :)
If only devs came back with the follow up that was promised...
http://www.eveonline.com/ingameboard.asp?a=topic&threadID=1489084
|
Zagdul
Gallente Clan Shadow Wolf Fatal Ascension
|
Posted - 2011.08.02 07:29:00 -
[94]
Edited by: Zagdul on 02/08/2011 07:33:37
Originally by: DeadlyThreat
Originally by: Zagdul The CPU's on tranq are more powerful PER CORE than the new Xeons.
Are you saying that the Wolfdale architecture is faster than the Westmere-EP architecture?
No.
Faster != better per core performance.
Yes, the new Xeons 32nm 10 core tech is gorgeous with 30mb of cache for multi tasking.
however, for single threaded tech and tasking the way python handles it, right now the only thing that can match it is the cell tech or new GPU's nvidia are putting out on a per core speed and gflops. Both would require a full code rewrite of the server engine.
I'm very sure people will correct me for being a moron here if I'm wrong, but this has been stated in multiple dev blogs as to why sometimes the "fast" processor doesn't translate to "best" for tranquility as the servers require a specific type of processor, not always the one that benchmarks nicely.
|
Salpun
Gallente Paramount Commerce
|
Posted - 2011.08.02 21:10:00 -
[95]
As soon as Bluenet gets applyed on the spaceship side some of the talking points about the unusablity of multicores might go out the window. I see some big wins in the future.
Then they can focus on Crime watch so that HS/LS will not grind to a halt when big fights happen.
|
Mollernak
Oberon Incorporated Morsus Mihi
|
Posted - 2011.08.05 01:38:00 -
[96]
That is a simply gorgeous tool..... I have had to look after some fairly complex systems in my time, and this sort of tool has always been my dream to get my hands on.
Very very jealous, but that said, happy it's you guys that have got your dirty little mitts on it.
A picture is worth a 1000 words, but a video is worth a book... Keep up this sort of "show and tell" blog, it's great for us to see what you guys are doing.
|
Vincent Jarjadian
|
Posted - 2011.08.09 20:37:00 -
[97]
I've been reading up on some of the technologies used atm to move virtual servers, such as Virtualbox's 'Teleport' feature... Supposedly they can move a VM between hosts while it is operating without disruption...
could this be used to remap lightly loaded systems off of heavily used nodes... might mean less horrific code, but also might mean more server OS licences required...
I know it's been said that lightly loaded systems dont affect the heavily loaded ones... but every little helps and is doesnt look that hard to do...
|
Kyros Emerson
Amarr Roving Guns Inc. RAZOR Alliance
|
Posted - 2011.08.26 06:31:00 -
[98]
Edited by: Kyros Emerson on 26/08/2011 06:33:06 Im not the most techno-savvy guy, but heres my question: You mentioned 'this is not a reenforced node', and that there were 1000+ people in the system. Has CCP considered making/inventing/theorycrafting a Dynamic Node Re-enforcement thing that detects when there are a large group of players in an area, and then reenforces said node automatically? Is that what the guy above me was saying about teleporting?
|
drinking12many
Minmatar Nine Inch Ninja Corp
|
Posted - 2011.08.26 15:18:00 -
[99]
Edited by: drinking12many on 26/08/2011 15:18:36
Originally by: Kyros Emerson Edited by: Kyros Emerson on 26/08/2011 06:33:06 Im not the most techno-savvy guy, but heres my question: You mentioned 'this is not a reenforced node', and that there were 1000+ people in the system. Has CCP considered making/inventing/theorycrafting a Dynamic Node Re-enforcement thing that detects when there are a large group of players in an area, and then reenforces said node automatically? Is that what the guy above me was saying about teleporting?
The problem with doing that was stated above to move it to another node to reinforce it would disconnect everyone in any other systems running on the same node when they move them.
|
K Kerryngktonn
|
Posted - 2011.08.27 21:41:00 -
[100]
That's some very nice tech pron indeed! I remember the first time I profiled my PHP code with a cache grind, I was happy as a pig, but this telemetry thing of yours is so detailed, visually appealing and intuitive after the initial "omg what's that?!" state! This tool should not only help fight with fleet lags, but presents a good opportunity to profile other game aspects as well, for which I'm very grateful :)
Hope you don't run out of space dumping all the telemetry data.
|
|
ShadowDncr01
|
Posted - 2011.09.05 04:29:00 -
[101]
Veritas,
Stupid question time, yah! I see that sending out the data to the interested parties takes a measuable amount time inside the fucntion "SingleCastBy ClientID". Have ya'll considered using PGM? Clients could subscribe to a multicast group which would might help offload some work from the server.
Also, any thoughts on using either numa or a fpga for the physics simulation code? Is there a reason why that can't be done in parallel with "UpdateClients" ?
|
|
|
|
Pages: 1 2 3 [4] :: one page |
First page | Previous page | Next page | Last page |