open All Channels
seplocked EVE Information Portal
blankseplocked New Dev Bog: Historical Server CPU Utilization Trend
 
This thread is older than 90 days and has been locked due to inactivity.


 
Pages: 1 2 [3]

Author Topic

iqplayer
Caldari
Dragon's Rage
Posted - 2010.11.17 16:08:00 - [61]
 

Quote:
I really wish we had a metric for "number of players that can all shoot each other while the server is nice and happy", but I'm not even sure how we'd go about defining that, let alone getting enough samples to make it a sane metric. It stands to reason though that if player actions are requiring less CPU in general, they should require less in a big fleet fight than before, which would yield more people with similar performance.



Assuming that CPU/user is gathered per server, what might tell you this is actually a discrepancy between the actual and average cpu usage per person. That is, normally a person engaged in active PvP should be using more CPU than the average user, so systems with high load AND high usage per person would likely mean active PvP (or possibly a popular PvE system). From the Blog a few days ago, though, once a node hits it's limit, it starts delaying tasks until a breakeven is reached. So, the more players in system, the lower the amount of CPU per person would be - so an overloaded system should show near 100% CPU usage AND a lower than average cpu/person.

No Pun
Poitot Intelligence Service
Posted - 2010.11.17 17:03:00 - [62]
 

Edited by: No Pun on 17/11/2010 17:05:22
Originally by: Gnulpie
Great blog! Really good stuff, thanks.

But I am confused by the many little spikes in the first graph. Why are there so many little spikes?

Why is the CPU per User going up noticable, just to go down a few day(s) later again? I would have expected the CPU per user being pretty continuous.


My quick guess would be; 4 times per month, so it's the weekend spikes (more players more usage). Didn't get averaged out, since data wasn't averaged in 1-week window.

The more sporadic spikes then are the interesting question; dev's want to throw some guesses what are the bigger spikes, especially after the Gridlock start?

Illectroculus Defined
No Bull Ships
Posted - 2010.11.17 17:33:00 - [63]
 

Regarding the Drake graph - I'm pretty sure one of the big catalysts driving a growth in drakes for PVP was the introduction of Sized rigs. This dropped the cost of losing a rigged drake (or other battlecruiser) by 30million isk overnight, and when you factor in the insurance the reduction in the PVP cost becomes a huge factor. Granted the graph shows a blip a couple of months before this change, but It's such a large boost to the cost/efficacy equation I can't imagine it's not an important driver.

Lykouleon
Wildly Inappropriate
Goonswarm Federation
Posted - 2010.11.17 18:36:00 - [64]
 

Came to this blog expecting a graph-gasm

Left moderately disappointed. WTF CCP, MOAR GRAPHS!!!!

And also pg 3 with a Chribba post? Neutral

Terra Mikael
Horizon Dynamics
Posted - 2010.11.18 02:35:00 - [65]
 

BECAUSE OF DRAKE!?

Sessym
Amarr
Posted - 2010.11.18 09:03:00 - [66]
 

It is interesting to see how small facelifts for your old game begins to take substantial effect. Please say that one day we may be rid of the all the stuff that goes nuts after an update hits. That'd be heaven Cool

Also, where is Chribba? Veldspar cries.

Grady Eltoren
Minmatar
Aviation Professionals for EVE
Posted - 2010.11.18 11:10:00 - [67]
 

Originally by: Deviana Sevidon
Active tanking in PvP stopped to a long time ago to be a viable alternative, so it was either pure damage or buffer tank. With more and more Titans appearing in the boosts to buffer tanks, the pure damage setup also fell mostly out of favor. After that happened the race was only between speed/blob and buffer/RR/blob.

It is not surprising that so many use the drake, buffer-armor tanks have the massive disadvantage of speed and mobility, so they are only viable if you have massive amounts of RR. Shield tanks are more forgiving, many player hardly notice the sig-radius increase and the fact that these shields have high amount of passive regeneration on top of their buffer also helps.

The problem is not the drake but passive tanking and especially shield tanking in general. It simply encouraging blobbing too much and brought to eve the cycle of ever increasing blobs. Blobs that the hardware cannot support. Even if CCPs makes drastic improvements to the hardware, the blobs would just increase again to the point where the hardware is again unable to support fleet battles.


THIS.

And expanding on the point of needing RR gangs for buffer tanks - you just don't see these as much due to lag so it is a vicious circle.

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2010.11.18 12:21:00 - [68]
 

Originally by: CCP Veritas
Originally by: Bartholomeus Crane
The underlying problem is not necessarily the ship used and it's CPU utilisation.


I agree with you, but my point in showing that graph isn't that fixing missiles will fix lag, it's that making them performant should help reverse some of the degradation we've seen in large fleet engagements around Dominion-launch-time. They really are quite exceptionally worse than guns, in ways that I'm quite certain would offend your sensibilities.

I'm well aware that all the optimizations we're doing at this point are, in the limit, futile, since we always allow/encourage players to bring one more ship. Since I don't know when/if that's going to change, I'll keep aiming down the path of supporting the current design as best we can.


And from your POV, and that of the Girdlock team, that is exactly what you should do, and continue to do. It might not mean much, but I think you guys are on the right track. It seems to me you are profiling the problem, identifying a part of the server that contributes the most to the problem, and try to speed it up, in a continuous improvement cycle. And the first graph suggests that this is working. Clearly, I'm not saying you should stop doing that. And if the missile handling is in comparison to turret handling (and/or other event handling) exceptional in contribution to the problem, as you say it is, then this should be addressed with priority. The same applies to fixing the O(n^2) subroutines in Destiny, with profiling data again providing guidance towards priority.

That this is futile in the end, which I agree it is, is by no means a reason to stop making improvements. Far from it in fact. However, I'd be very hesitant to use profiling data generated to improve the server performance to inform game design changes. And this is suggested by including the second graph.

Although there should be a feedback loop between the Girdlock team and the game designers, in the case of the nerf-Drake narrative CCP seems to be building, I think the wrong conclusions are being drawn. If the Drake is to be nerfed (which I think in this case would simply be combating a symptom), it shouldn't be because it causes problems for the server as a result of an artefact there (inefficient missile handling). I don't think you should game design a problem away when it is ultimately a problem that needs a technical solution.

In the end I don't think that Drakes in this case are overpowered. They are an easy ship to get into, but they are not vastly better than other ships in the game. Ultimately the underlying problem is one of blobs. And if game design can find an answer for that, all the efforts by team Gridlock will become less futile. Drakes themselves aren't a problem in and off themselves, blobs are!

Bartholomeus Crane
Gallente
The Crane Family
Posted - 2010.11.18 12:42:00 - [69]
 

Edited by: Bartholomeus Crane on 18/11/2010 12:42:22
Originally by: Marchocias
Originally by: Bartholomeus Crane
It's easy to end up in a race-condition that way.


Some good discussion dude, but I don't think "race condition" means what you think it means. I reckon you mean arms-race.


Quote:

A race-condition or race-hazard is a flaw in a system or process whereby the output and/or result of the process is unexpectedly and critically dependent on the sequence or timing of other events. The term originates with the idea of two events racing each other to influence the output first.



The system in this case is EVE. The output is the performance of the server, in this case, the amount of lag. One event-sequence is the players choosing their FOTM ship. Another event-sequence is CCP nerfing ships because their effect on server performance (or simply because they are FOTM). The two event sequences race each other to influence the server performance (although with different objectives). The application isn't exact, but it's close enough. CCP shouldn't get into a nerfing race with the players based on server performance. In the end you might end up nerfing all ships that happen to have been popular. Especially not when there's an underlying problem whose solution would render the whole exercise moot anyway: blobbing.

I don't think arms-race applies because although there is a tit-for-tat reaction to it, there isn't the progressive build up of 'arms'. Quite the contrary in fact.

vipeer
D00M.
Northern Coalition.
Posted - 2010.11.20 11:39:00 - [70]
 

We use drakes because they are sooooo much better than any other Battlecruiser it's not even funny.

There are a few types of viable gangs for pvp? LR or Nano HAC's or RR BS's or drakes+scimis

Therein comes the "problem" of drakes. Those ships by far the cheapest choice.

Obvious resolution of this problem would be to nerf drakes or boost all other BC's. I'll go with the easier of two: Swap their shield resistances bonus for something else or lower the CPU so you make ppl fit a processor or make drake pilots use close range missiles via lowering targetting range...

And BTW...You guys at CCP ever think about why nobody in their right mind uses blasters anymore and how to fix that?

Black Dranzer
Caldari
Posted - 2010.11.20 14:15:00 - [71]
 

Wait, something's off here. Why would nuking a bunch of bots reduce the CPU per user? Unless a bot uses up considerably more server side CPU than your average pilot, there's going to be a drop in the overall, but not the per user. And if bots do use more server CPU than everybody else.. well, that's pretty ****ing baffling to be honest.

I suppose it could be explained if you're taking the CPU use of idle docked players into account, but then you're not graphing CPU use, you're graphing average player activity.

Mikel Laurentson
Posted - 2010.11.20 15:05:00 - [72]
 

Originally by: Black Dranzer
Wait, something's off here. Why would nuking a bunch of bots reduce the CPU per user? Unless a bot uses up considerably more server side CPU than your average pilot, there's going to be a drop in the overall, but not the per user. And if bots do use more server CPU than everybody else.. well, that's pretty ****ing baffling to be honest.

I suppose it could be explained if you're taking the CPU use of idle docked players into account, but then you're not graphing CPU use, you're graphing average player activity.


What makes you think that the 'idle docked' players don't require CPU resources? People playing the market, fiddling with manufacturing, or just using the channels are all asking the servers to do stuff.

Dr Cron
Northern Lights Number 5
Hydroponic Zone
Posted - 2010.11.20 15:31:00 - [73]
 

I'm not getting any useful information out of this graph.

"We'll just conveniently not address that HUGE second spike" - CCP

Black Dranzer
Caldari
Posted - 2010.11.20 18:41:00 - [74]
 

Originally by: Mikel Laurentson
What makes you think that the 'idle docked' players don't require CPU resources? People playing the market, fiddling with manufacturing, or just using the channels are all asking the servers to do stuff.

It's not that; It's just that it's not the best for measuring real performance. A drop in the graph could indicate a performance increase, but it could also indicate people sitting in their stations more. In this case it probably doesn't generate too much noise, but that doesn't make it a good metric.

Gnulpie
Minmatar
Miner Tech
Posted - 2010.11.22 09:59:00 - [75]
 

Originally by: Black Dranzer
Wait, something's off here. Why would nuking a bunch of bots reduce the CPU per user? Unless a bot uses up considerably more server side CPU than your average pilot...


Exactly.

The bots used up a HUGE amount of CPU by spamming all the time the directional scanner (back at that time there was no cooldown counter for the scanner), several times a second ALL THE TIME, to scan for hostile ships on the directional scanner and at first sight of them on they logged off. This consumed a lot of CPU. A lot more than the average user used.



Drakes?

Drakes aren't used because they are so great but because they work in lag better than other ships! If there wouldn't be lag, drakes wouldn't be used in such quantities.


chatgris
Quantum Cats Syndicate
Posted - 2010.11.22 12:57:00 - [76]
 

Quote:
Dominion 1.0 was a bit worse than Apocrypha 1.5 for load. Not hugely so however


How is this the case when the Dominion 1.0 has a lower CPU per user metric? It looks to me that this comment is incorrect.

Junkie Babe
Posted - 2010.11.22 20:50:00 - [77]
 

Edited by: Junkie Babe on 22/11/2010 20:51:01
Originally by: Antihrist Pripravnik
Edited by: Antihrist Pripravnik on 17/11/2010 07:27:59
Originally by: Lili Lu
Baseless blabbering about Drakes


Yeah, sure. You are an expert on Drakes right? The biggest experts don't ever fly ships they criticize.Rolling Eyes

Drake is the only Caldari BC that uses missiles (unlike Amarr or Minmatar that use their primary weapon type on both BCs). Drake can't active tank (unlike Amarr or Minmatar BC's). Unlike you, who are crosstrained between Amarr and Minmatar, Caldari missile pilots do not have a choice which ships they are going to fly - 1 cruiser, 1 BC, 1 BS, 1 HAC (if you can call it that way... don't know what "Heavy" stands for in Cerberus's name anyway). Having a large percent of players in game in Caldari race and missile trained, you are basically asking that they nerf the only good thing that Drake have - its tank (which is btw not anywhere near Cyclone's active tank in 1v1... but who cares, right?Rolling Eyes)




The only thing a drake has is it's tank?

Drake buffer fit

http://eve.battleclinic.com/loadout/44298-Command-Ham-Drake.html

PVP Harby fit
[Harbinger, PVP]
Heat Sink II
Heat Sink II
Damage Control II
Energized Adaptive Nano Membrane II
Energized Adaptive Nano Membrane II
1600mm Reinforced Rolled Tungsten Plates I

10MN Afterburner II
X5 Prototype I Engine Enervator
X5 Prototype I Engine Enervator
Warp Scrambler II

Focused Medium Pulse Laser II, Scorch M
Focused Medium Pulse Laser II, Scorch M
Focused Medium Pulse Laser II, Scorch M
Focused Medium Pulse Laser II, Scorch M
Focused Medium Pulse Laser II, Scorch M
Focused Medium Pulse Laser II, Scorch M
Focused Medium Pulse Laser II, Scorch M
Medium Energy Neutralizer II

Medium Trimark Armor Pump I
Medium Trimark Armor Pump I
Medium Trimark Armor Pump I

Buffer tank
Drake 89k
Harby 76k

DPS Overloaded without drones
Drake 604 @ 18.1 KM with Terror Rage Assault missile
Harby 424 @ 20km with 3.8KM falloff with Scorch

So your drake has both a 13k EFHP bonus and does 180 more DPS.

Not that I'm saying all ships should be equal but doing the math,

Harby EFHP 76,000 / 604 Drake DPS is 126 seconds to destroy the Harby during which the Harby would have done 53,350 to the Drake leaving the Drake with over 35k EFHP left after the Harby had blown up.

As you can see the Drake is clearly overpowered it is a mile better than the Harby.




Vidar Kentoran
Minmatar
Eighty Joule Brewery
Posted - 2010.11.23 00:06:00 - [78]
 

Originally by: Junkie Babe

As you can see the Drake is clearly overpowered it is a mile better than the Harby.



You're basically ignoring the relative effectiveness of the weapon systems you're talking about. Just because a Drake can 1v1 a Harbinger doesn't mean anything.

Now, if you could prove the Drake was more effective against every target in the game than the Harbinger, you might be correct. (Hint: it's not)

Rip Minner
Gallente
ARMITAGE Logistics Salvage and Industries
Posted - 2010.11.23 06:35:00 - [79]
 

Love the graph chart.

<--- Unholy rage. See numbers do count. Twisted Evil

Rylai
Gallente
Reikoku
IT Alliance
Posted - 2010.11.25 20:48:00 - [80]
 

OMG GET RID OF MISSILES

and happy thanksgiving (2nd international holiday ever)(after fourth of july)


Pages: 1 2 [3]

This thread is older than 90 days and has been locked due to inactivity.


 


The new forums are live

Please adjust your bookmarks to https://forums.eveonline.com

These forums are archived and read-only