Onictus
Silver Snake Enterprise
Fatal Ascension
601
   |
Posted - 2013.11.24 13:10:00 -
[1] - Quote
Decian Cor wrote:Frostys Virpio wrote:Decian Cor wrote:Maybe I am just terribly naive, but couldn't this all be resolved with more/better hardware that was able to deal these kinds of fights?
I'm sure there is money for it. The best node CCP have on hands right now were described by a DEV as OC'd Xeon CPU cores. At some point, you can;t really push the clocks any higher while keeping the system stable for long enough. LN2 cooling is cool and all to set records but you can;t really run **** for full days with that. Having a server admin being paid to pour LN2 in the CPU cooler all day? Definitely gonna happen... NOT. I'm not too savvy with computers, so I figured it was worth the question. You say it is unrealistic to have it run like that for days on end...would it be possible to just have something like that kick in when major activities such as these battles happen? then it'd be a max of maybe 6 hours at a time, whenever there was a huge fight?
There are a couple issues.
Its not hardware. CCP has very very top shelf servers the X4000 series they use for the Jita node is up there with the best enterprise level servers on the market.
The issue is the Python, Python is single thread, and since newer server architecture uses more cores to spread a hyperthreaded application and get more done at once, this isn't possible on Eve because Python is single thread, you really can't multi-thread. The other thing would be dynamic load balancing i.e. auto reinforcement, the problem there is that in a real time (computer real time) environment that is VERY VERY difficult, my company was working on it for like 4-5 years and basically gave up, again, a single thread application real times.
Since porting Eve to a multi-threaded program would basically mean re-writing the entire game from the bottom up this is a non-trival task and would best be dubbed "Eve Onine 2" so they are working with things like brain in a box and inifiniband to try to reduce the overall load.
Still there is no simple "fix" everytime they increase cluster perfomance we just jam more people in system. |