|
|
| Author |
Thread Statistics | Show CCP posts - 10 post(s) |
|
CCP Fallout
   |
Posted - 2010.09.13 16:53:00 -
[1]
CCP Veritas' new "Fixing Lag" blog focuses on player module repeat optimization. You can read all about it here.
Fallout
Associate Community Manager
CCP Hf, EVE Online
Contact us |
|
|
CCP Explorer
|
Posted - 2010.09.13 17:10:00 -
[2]
 Originally by: London
IBC, and awesome job. Does this only apply to Civilian guns for now? I understand it's a stepping stone towards other module optimization.
This applies to any module that is repeating.
Civilian modules were used in the in-lab tests simply to make sure that inventory operations on ammo didn't introduce any noise in the measurements.
Erlendur S. Thorsteinsson
Software Director
EVE Online, CCP Games |
|
|
CCP Veritas
|
Posted - 2010.09.13 17:21:00 -
[3]
Originally by: Liang Nuren
I'm loving the amount of new blogs... I hope CCP keeps going with them
I for one certainly plan to. The report structure of this is something we've been doing internally just for our own bookkeeping and peer review purposes. Converting them to devblogs is fairly painless. I'm not saying all of them are going to hit devblog, but the big ones should.
|
|
|
CCP Veritas
|
Posted - 2010.09.13 17:37:00 -
[4]
Originally by: dangerfan
I have a question about the general cpu usage graph at the end. The graph's sample size shows there are 2 down-times before the flag, and 4 down-times after the flag. Was this taken into consideration when calculating the average cpu usage, as to not skew the results?
Don't take that graph too seriously. CPU usage per player is a pretty squishy metric to begin with, and no two days in Eve are the same. If I could take 15 days with the change and 15 without, then compare the numbers for those in aggregate, it would be much more meaningful. Of course, that would mean 30 days of not being able to do anything else to the cluster, which is not a cost I'm willing to pay.
The take-away from the last chart isn't that it was an exact 5% drop (that number came from eyeballing the graph, so, yeah, really weak), but rather that a cluster-wide effect was felt. For such a trivial change, that's awesome. Well, 8.5 awesomes, to be exact.
|
|
|
CCP Veritas
|
Posted - 2010.09.13 20:36:00 -
[5]
Originally by: Charles37
I'll read the rest of the blog after I've stripped my Maelstrom of it's current setup in favor of one the CCP sanctioned setup that uses all Civilian modules.
I'm tellin' ya, it can kill a Titan! Well, so long as the Titan isn't piloted and you have a lot of time.
|
|
|
CCP Explorer
|
Posted - 2010.09.13 22:50:00 -
[6]
Originally by: Nye Jaran
Good blog, but your math is wrong here:
Quote:
A drop from approx 82% CPU to approx 35%, a 57% improvement
Should be a 47% improvement, or the pre /post number(s) are wrong.
82% to 35% is an improvement of 47 percentage points but 1-35/82 is 0.57 or 57%.
Erlendur S. Thorsteinsson
Software Director
EVE Online, CCP Games |
|
|
CCP Veritas
|
Posted - 2010.09.13 22:52:00 -
[7]
Damn it Explorer, beat me to it!
But yeah, what he said. Relative difference is what matters, not absolute.
|
|
|
CCP Explorer
|
Posted - 2010.09.14 08:29:00 -
[8]
Originally by: Louis deGuerre
Nice work. Did you get same results using missiles and guns or is that just an assumption ?
All repeating modules.
Erlendur S. Thorsteinsson
Software Director
EVE Online, CCP Games |
|
|
CCP Veritas
|
Posted - 2010.09.14 17:42:00 -
[9]
Originally by: Trebor Daehdoow
Since this is the case, would it not be a good idea, as several people have suggested, to consider making some common modules passive -- such as Damage/Drone Control Units?
I don't know how much of a win it would be offhand, but I have been keeping an eye on your thread over in the Assembly about it. Very interesting discussion, that.
|
|
|
CCP Veritas
|
Posted - 2010.09.16 17:30:00 -
[10]
Originally by: Andrevv
Seeing as 0.0 combat involves manual cycling anyway, I don't expect this particular optimization to have any impact on fleet fights. Good job regardless :)
I had a look at the code again and wish to issue a correction. This check happens on every module activation or repeat, so it will impact manual cycling just the same.
Originally by: Andrevv
Am I understanding this correctly, that any repeating module, regardless of whether it takes a charge, does a query to check if its charge group is empty?? because that is what the blog is implying... which seems kind of pointless.
Yes, you are understanding correctly. Thankfully, after this change, it's almost certainly cheaper to ask "Is this reloading?" than "Can this have charges? If so, is it reloading?"
|
|
|
|
|
|
| |
|