These forums have been archived and are now read-only.

The new forums are live and can be found at https://forums.eveonline.com/

EVE Information Portal

 
  • Topic is locked indefinitely.
Previous page123Next page
 

New dev blog: Tama Dev Event - The Tech Side

First post
Author
Previous Topic Next Topic
Max Kolonko
Caldari Provisions
Caldari State
#21 - 2011-12-16 20:30:20 UTC
maybe next itme you run this event it will be with TD on?

Read and support: Don't mess with OUR WH's What is Your stance on WH stuff?
CCP Veritas
C C P
C C P Alliance
#22 - 2011-12-16 20:30:31 UTC
Whiteknight03 wrote:
So did time dilation go into effect during this fight, and if so, to what degree/percentage?

Also, very nice

Time Dilation has not yet been activated on TQ due to some issues found in late testing. Had it been on, it would have kicked in during the last fleet warpin, down to 80% of normal time or so, and then come back up to normal.

CCP Veritas - Technical Director - EVE Online
Boeboe Joe
Royal Assassins Guild
Chained Reactions
#23 - 2011-12-16 20:39:49 UTC
Great dev blog CCP! I especially like the use of graphs and simple-ish explaining of what makes those hamst...err I mean hardware upgrades do for our game. Can't wait to see what y'all have in store for TQ next year!

Got 'roids?
FlinchingNinja Kishunuba
Crunchy Crunchy
#24 - 2011-12-16 20:44:58 UTC
There are a few companies offering OC servers primarily for stock market companies.

Below article talks about one,
http://www.theregister.co.uk/2010/12/17/sgi_rackable_overclocked_server/

You guys better sort out your multi-threading though your not going to get better clocks any time soon. You may see a 5GHz part in the next few years as a dual core (huge maybe). But that's about if for X86 server parts.
Rixiu
PonyTek
#25 - 2011-12-16 21:11:10 UTC
Nice to see some client improvements, especially GPU load could use some load.

FlinchingNinja Kishunuba wrote:
There are a few companies offering OC servers primarily for stock market companies.

Below article talks about one,
http://www.theregister.co.uk/2010/12/17/sgi_rackable_overclocked_server/

You guys better sort out your multi-threading though your not going to get better clocks any time soon. You may see a 5GHz part in the next few years as a dual core (huge maybe). But that's about if for X86 server parts.


"Moores law" (not really a law but oh well) will still apply for some more time so the performance per clock will continue to increase and we'll get performance that way. CPU clocks have been more or less the same ever since intels netburst architecture >5 years ago and CPU performance are hardly the same as back then.
Kyoko Sakoda
Achura-Waschi Exchange
Monyusaiya Industry Trade Group
#26 - 2011-12-16 21:43:10 UTC
I understand it's a pretty big project to undertake, but will the server code ever be optimized for multi-core processing?
CCP Veritas
C C P
C C P Alliance
#27 - 2011-12-16 22:04:54 UTC
Kyoko Sakoda wrote:
I understand it's a pretty big project to undertake, but will the server code ever be optimized for multi-core processing?

Yep. There's a plan on the table for a pilot project taking an identified piece of heavy load off to a second independent thread. We need to prioritize that versus other things like client performance and other possible server gains, but there's plans at least.

CCP Veritas - Technical Director - EVE Online
Lowa
North Star Networks
Not Purple Shoot It.
#28 - 2011-12-16 22:08:32 UTC
So I wrote this nice post full of love and devotion but your balloon animal of a forum ate it so this will be the short version:

GPU and DSP compute power as seen in the HPC space, is that anything you are looking into?



The remedial task of porting the code over to test I leave to you, I mean come on, how hard can it be?
Daelorn
State War Academy
Caldari State
#29 - 2011-12-16 22:12:12 UTC
hired goon wrote:
Wow, the mention of client-side performance enhancements is really interesting and exciting. I have always been puzzled at Eve's scaling problems; how it performs so well on even lowly hardware but suddenly falls to it's knees when there is a large fleet on the field. I always imagined it'd be as simple as loading simpler ship models/lower resolution textures when there are many ships on screen.... but it's never as simple as we think!

Thanks and great devblog Smile


No I'm by no means an expert in this but rending of ships, and stations is purely done on the client. This has nothing to do with the server. Having lower/higher poly models won't help or worsen the server. The server is trying to coordinate who is getting show, how much, where they are in space, how fast they are moving and such. Add in drones, missiles, repairing, aggression, things like that times 800 people and the server begins to cry.
Jack bubu
Caldari Provisions
Caldari State
#30 - 2011-12-16 22:24:46 UTC
Daelorn wrote:
hired goon wrote:
Wow, the mention of client-side performance enhancements is really interesting and exciting. I have always been puzzled at Eve's scaling problems; how it performs so well on even lowly hardware but suddenly falls to it's knees when there is a large fleet on the field. I always imagined it'd be as simple as loading simpler ship models/lower resolution textures when there are many ships on screen.... but it's never as simple as we think!

Thanks and great devblog Smile


No I'm by no means an expert in this but rending of ships, and stations is purely done on the client. This has nothing to do with the server. Having lower/higher poly models won't help or worsen the server. The server is trying to coordinate who is getting show, how much, where they are in space, how fast they are moving and such. Add in drones, missiles, repairing, aggression, things like that times 800 people and the server begins to cry.

He was talking about client-side performance
Darth Sith
Genbuku.
Psycho Unicorn Squad
#31 - 2011-12-16 22:44:12 UTC  |  Edited by: Darth Sith
Well, at the time that CCP was discussing it at Fanfest last year, Intel had this system under really tight raps.

Recently however they have relaxed the NDA and we are now allowed to publicly disclose the details on the box in general terms.

Here is the box in question : Monster Toy

and it is making use of Intel's beast of a proc running at 4.4 Ghz (air cooled) aka : Intel Xeon X5698 4.4 GHz


Take a wild guess who I work for :)

Cheers.
H3llHound
Center for Advanced Studies
Gallente Federation
#32 - 2011-12-17 00:23:20 UTC
Nice read but can you make(and show us :D) a server load graph of the fight in L-6BE1 on the 21.11.2011 from 22:00 to 24:00? Afaik that system wasnt reinforced and I would be interested how the servers kept up as it was lagging qiuet a bit for some time. pretty please :)

P.s. I will give you a 'like' if this happens ;)
Mioelnir
Brutor Tribe
Minmatar Republic
#33 - 2011-12-17 01:35:31 UTC
So, looking at the tama graph... its running on the supernode with +80% over regular dedicated nodes.

Which would put a regular node's 100% mark at around 55% of the supernode graph. So, had this not been on the node it was, there would have been constant 100% utilization for an hour between 19:30 and 20:30?
Klandi
Consortium of stella Technologies
#34 - 2011-12-17 01:46:00 UTC
Great info - thanks for the update

Of interest to me - how much do you consider is the lag apportioned to the server hardware vs the client hardware? Ultimately if we had the best hardware on the market for the client with your current server configuration then what would be achievable in a fight. 1000 each side in system?

I won't hold you to any guesses :)

I am aware of my own ignorance and have checked my emotional quotient - thanks for asking
Rognin
Space-Hogs
#35 - 2011-12-17 02:06:22 UTC  |  Edited by: Rognin
LOL EKWB

Seriously, you guys went about that one wrong...


Edit: And Liquid was DI water + PT nuke, and if it weren't for EK, then a silver kill coil would have done the job.

At least put that old hardware to folding?

Phase change or a TEC WB would have done a better job at cooling than your air solution.
ibnubis
School of Applied Knowledge
Caldari State
#36 - 2011-12-17 03:20:24 UTC
You guys are doing some great tuning. Very Impressive.

Have you guys also bumped the network along with the processor technology? If your servers get chatty to each other under load and you're using Ethernet, the TCP stack will load up your cores especially with lots of processes running. Or do you use InfiniBand? IB will scale linearly with core count and has no TCP stack. Very low latency, (700nS) and 56Gb/sec. IB can sometimes solve CPU headroom and scaling issues. As long as the application is dependent on server to server communication.
Sinq Arnolles
Center for Advanced Studies
Gallente Federation
#37 - 2011-12-17 04:57:52 UTC
2 fleets fights I have been in recently have lagged like crazy.

The worst one was http://eve-kill.net/?a=kill_related&kll_id=11742712
As a logistics pilot it took about a min to lock anything and a good 30 seconds to activate / deactivate a module and when the FC did a fleet warp it took 10 mins for it to actually start doing it. Sometimes your ship would break orbit and go in a random direction and it would just keep going for a few mins before it responded to your commands to go somewhere. Probing also took at-least 1 min to work.

http://eve-kill.net/?a=kill_related&kll_id=11762630 Similar situation for this one but it was slightly better.. still pretty impossible for logi to do anything though.
Aineko Macx
#38 - 2011-12-17 07:17:08 UTC
How is the fundamental research Jacky (CCP Warlock) is doing into Python multithreading, the GIL, etc. coming along?

Btw, I approve of client side performance improvements. Get them brackets performance fixed plz!

iveeCore 3.0: The PHP engine for industrial activities and CREST library
Darth Vapour
School of Applied Knowledge
Caldari State
#39 - 2011-12-17 08:41:44 UTC
Aineko Macx wrote:
How is the fundamental research Jacky (CCP Warlock) is doing into Python multithreading, the GIL, etc. coming along?



According to her LinkedIn profile CCP Warlock no longer works at CCP.
Camios
Sebiestor Tribe
Minmatar Republic
#40 - 2011-12-17 09:24:13 UTC  |  Edited by: Camios
These are awesome news (this is the meaning of awesome).
I have some questions:

How much does the Frankenstein server cost, compared to a "normal" one? That is, how many such blades can we expect to have in place on tranquillity in the future, considering that TiDi will greatly reduce the impact of "lag" on game mechanics thus reducing the need of extreme performance hardware?

Slightly out of topic:

Is there any room for improvements in the jump and dock-undock procedure? In the event on dec the 8th TiDi reached the 10% limit (is this correct?) during a mass jump and a mass dock. And it seems that the load generated by these actions is far heavier than the one generated by usual fighting, even missile spam (right?). This could theoretically be exploited by a side in combat since a side with lots of ships could, by keeping jumping in and out of a system, resurrect the old lag monster.

In the worst case, how many jumps can a node handle per second without lagging behind without time dilating? And how many warpins? And is it correct that with TiDi at 10% that number will be multiplied by 10? With these numbers and considering the duration of the (dilated) session change timer one can estimate how many people are necessary to do the exploit I described above.
Previous page123Next page