| Pages: 1 2 [3] :: one page |
| Author |
Thread Statistics | Show CCP posts - 0 post(s) |
Zafon
Caldari
FATAL REVELATIONS
Lotka Volterra
   |
Posted - 2006.10.11 12:08:00 -
[61]
Meh. I should have just said "CCP might not have the best source control available, that concerns me".
|
Semkhet
The Priory
Interstellar Alcohol Conglomerate
|
Posted - 2006.10.11 13:18:00 -
[62]
@Wilfan Ret'nub:
I don't intend to be drawn into a highly technical debate that would be obscure to every non-initiated in the intricacies of how OS / API's and networked applications interact in a distributed environment.
However, I did'n make a general statement as you implied but directly compared Python to other languages. And if I did not mention specifically "Stackless Python" as you where so keen to point out, it is for a very simple reason: none of the fundamental aspects that render Python compilers unable to generate very optimized code are adressed in the "stackless" version.
I cite the own words of Stackless developpers:
"Stackless Python is an enhanced version of the Python programming language. It allows programmers to reap the benefits of thread-based programming without the performance and complexity problems associated with conventional threads. The microthreads that Stackless adds to Python are a cheap and lightweight convenience which can if used properly, give the following benefits:
+ Improved program structure.
+ More readable code.
+ Increased programmer productivity.
"
Again, it's very nice on the paper concerning the modelisation and design phase, and may improve the amount of code produced in a given time. But where go reliability, fault tolerance and debugging quality at MACHINE CODE ? Because I hope you are aware that at the end, no matter which coding strategy you use, its' still conventional machine code that CPU's will swallow, right ?
And here we reach one of the biggest paradox of modern IT technologies: more ergonomic the tools (understand closer to classic human reasoning), faster you can program, but more difficult gets the debugging since each superposed layer of code acts like an additional black box.
Now a formal example of a fundamental aspect of Python, which is very convenient when programming, but renders optimisation difficult not to say hardly faisible.
In Python, you can write:
MyVar = 2006
MyVar = "This is a string"
These statements reference first an integer object, then a string object. There was no need to formally declare MyVAR and assign a fixed data type before using it.
This implies that because the type of the variable can change at any moment in time, Python compilers usually don't track variable assignements hence types, and handle the variable as a generic data object. Under such strategy, even simple mathematical operations require an extensive array of intermediate transformation and checking code just to know if said mathematical operation is pertinent.
Compare that to stricly static-typed languages like C or assembler, where a variable is defined with its type, which automatically defines BEFORE COMPILATION the required memory footprint and limits the range of operations which can be applied, and allows the compiler to detect and notify any trivial illegal use of said variables.
As far as I know, Python doesen't define any syntax for type declaration. Also, type inference could theorically solve this problem, but would crash frontally with the dynamic spirit on which Python is based. Basic line is, you can't have the best of those two worlds at the same time because the implications are mutually exclusive.
Finally, I spoke about a client written in a mix of C and inline assembly. I don't know where you get your opinions from, but I can assure you that something that will never become obsolete is the ability to fine-tune programs by resorting to the least time-consuming code. This is achieved by writing critical sections in low-level languages. Maybe you believe that BIOS'es, OS and their API's + drivers are coded in Delphi ? Check Linux: C is still one of the kings.
I agree with you that no amount of lag should crash a node. But here we go: further you go away from low level, and less are you able to understand what happens deep inside your creation on layers you don't master.
|
Semkhet
The Priory
Interstellar Alcohol Conglomerate
|
Posted - 2006.10.11 13:47:00 -
[63]
 Originally by: Nez Perces
did you write all that out yourself? cause if you did.. thx  ... good reading.... even if you didnt it was still a very good read..
anyways.. If what you are saying is true, then CCP's hands may be pretty tied.
Thx m8. IMHO it will take an horrendous amount of time to CCP to figure out where are the errors. I can cite the follow real-life example of a stupid problem I was once confronted, hence my understanding towards the poor folks at CCP.
One day I get a project to develop a distributed system of automation controllers in a highly polluted EM industrial environment. We choose the hardware (Siemens Infineon microcontrollers), design the missing electronic interfaces, write the BIOS and the application. And like everybody else, since I don't like to rediscover the wheel I resort to a third party for a library covering the low-level networking routines and the corresponding hardware network extension.
The day the system is fired up, everything works perfectly. Then we begin to add features. And comes a point where the networking becomes erratic despite the controllers enjoying a decent CPU load reserve. Simply put, the network acts like if someone is plugging and unplugging the connection cables randomly.
After WEEKS of litteral headaches where we can't find any bug in our code (ANSI C + assembler), we decide to reverse-engineer the network library and begin to map the machine code to the schematics of the processor and the hardware network extension.
And what do we find ? The network board included an RTC chip that we found very handy to implement concurrent timing routines. However, this chip had one of its UNUSED digital output line going high (+5V) at regular intervals. Since we did'n tie this line to anything we initially thought that there was nothing to investigate in this direction. But at the end, by careful examination of the schematics, we discovered that this line was INTERNALLY connected to a component involved in the network reception hardware, preventing said component to work as intended during small delays of a few milliseconds.
This was the bug. The company which developped the network board never realized that there was a precise functioning mode of the RTC chip which was incompatible with the concurrent network reception ability because none of their clients had used that special RTC mode.
As conclusion, that's why I always adopt caution in development and refrain to make inflammatory statements. We often don't realize how far and deep things can get screwed up in the most unexpected way. And if I would have been one of those who rely solely on librairies, without a solid knowledge of digital electronics, processors and assembler, we probably never have been able to find and correct this bug ourselves.
|
Silvestri
FIRMA
Interstellar Alcohol Conglomerate
|
Posted - 2006.10.11 14:11:00 -
[64]
Maybe Eve should check out my company...lol.
Avid Technology
Specialize in editing equip for movies...
With our ISIS storage and bazillion different servers...maybe there's a way...lol.
16 Terrabyte chassis's.....and you can hook them up for more....I think a total of 8 chassis's...
|
NATMav
F.R.E.E. Explorer
EVE Animal Control
|
Posted - 2006.10.11 15:08:00 -
[65]
It's getting to the point where there really needs to be some drastic action taken to try to remedy the problem, whether it be disabling entire features, such as bookmarks, setting a maximum number of players per node, removing POS, or whatever it takes to track down the source of the problem.
Yes, it will be difficult and painful, and there will be a lot of whining from the playerbase, but if CCP comes out up-front and says, "This week we're taking out bookmarks to see how it affects server performance. At the end of this test all bookmarks will be returned in their original state.", the majority of the playerbase will deal with it maturely, and if it helps track down the source of these problems, we'll all be alot happier.
We know that they are working on tracking down the problems on Singularity, and have reproduced some of the issues, so hopefully that leads to the source of the problems. But if that is not the case, I think they have to take a real proactive approach and start eliminating potential sources one by one until it reveals itself.
FREEE is Recruiting |
Corsair Thunder
116.7
|
Posted - 2006.10.11 16:06:00 -
[66]
What could happen due to the node crashes, etc:
The large alliances, could split down into smaller ones, each controlling a small area of 0.0 space.
However, as people have said before, during the previous large-scale wars (GNW, Xetic & Stain Vs. Curse alliance) there was not a great deal of lag or server problems.
It is probably due to the increased number of effects on screen, plus the addition of capital ships, etc.
My 2 cents anyhow...
-----------------------------------------------
|
Serapis Aote
TBC
|
Posted - 2006.10.11 16:33:00 -
[67]
Originally by: Corsair Thunder
What could happen due to the node crashes, etc:
The large alliances, could split down into smaller ones, each controlling a small area of 0.0 space.
However, as people have said before, during the previous large-scale wars (GNW, Xetic & Stain Vs. Curse alliance) there was not a great deal of lag or server problems.
It is probably due to the increased number of effects on screen, plus the addition of capital ships, etc.
My 2 cents anyhow...
There are also a lot more people in Eve. then during those earlier wars. What was the average people on at peak during those wars. I remember when i started is was around 11k and that was after the GNW, now its hovers around 20k
I think the devs have stated that often low pop systems are put on the same node as high pop systems to balance things out. So while it may seem like lag in a 25-30 man fight is ridiculous, we have no idea how many other players on on the node that system is on.
Its hard to tell if its to many people in eve, bad code or what not...because we have had several patches and also a dramatic rise in players at roughly the same time.
|
Witch Doctor
Contraband Inc.
Mercenary Coalition
|
Posted - 2006.10.11 17:06:00 -
[68]
So ...
When is it getting fixed? That's all I care about.
|
Murukan
Minmatar
The Priory
Interstellar Alcohol Conglomerate
|
Posted - 2006.10.11 17:08:00 -
[69]
Originally by: Witch Doctor
So ...
When is it getting fixed? That's all I care about.
when parm stops wearing women's clothes 
In rust we trust!!! |
travelingsales
Ars Caelestis
Ascendant Frontier
|
Posted - 2006.10.11 17:09:00 -
[70]
Silvestri that's funny, I interned at Avid for a while. My desktop machine during the internship was a dual Xeon workstation with 1 terrabyte of attached RAID storage, they definitely don't skimp there 
To get back to the timeline thing a bit, I do not think "more players" is the problem causing the node deaths. If we think back to March or April there were frequently 20k+ players logged in and I do not recall hearing widespread reports of nodes dying. Granted the playerbase has grown further since then, but the average amount of players logged in at a given time is still around 20-23k.
To be honest the first couple of times I remember hearing about large numbers of node deaths were during the Goons vs. D2 war in the north. How long ago was that? I seem to recall it was right around when Dragon came out, which would fit with the general theory that something went wrong with Dragon. Before that there were various accusations thrown around during the LV vs. RA war (and let me stress that I am not accusing anyone of anything, just trying to write the "history of lag and node deaths") but during that war it seemed the problem was more individual players CTD'ing near POS, as opposed to whole nodes dropping and hundreds of players getting kicked at once.
Can anyone with a better memory than me confirm the timeline?
|
maGz
The Priory
Interstellar Alcohol Conglomerate
|
Posted - 2006.10.11 17:11:00 -
[71]
Originally by: Murukan
Originally by: Witch Doctor
So ...
When is it getting fixed? That's all I care about.
when parm stops wearing women's clothes
So...
never?
____________
The Priory Killboard |
Nez Perces
Amarr
Black Spot.
|
Posted - 2006.10.11 17:14:00 -
[72]
Originally by: travelingsales
To be honest the first couple of times I remember hearing about large numbers of node deaths were during the Goons vs. D2 war in the north. How long ago was that? I seem to recall it was right around when Dragon came out, which would fit with the general theory that something went wrong with Dragon.
Can anyone with a better memory than me confirm the timeline?
Actually the dragon code was introduced during the Goons vs D2 war (could be wrong .. but dont think so)... I do remember some D2 members hoping that the dragon code would solve their problems. The goon vs D2 war was plagued by problems from the start and there was no dragon code when it kicked off... ( i think..)
The main reason for the horrific problems with Goonfleet and D2 was the massive numbers GF were fielding.
Would be nice to hear more from D2 or GF about this.
|
Cicilus Hadrican
Caldari
Infinity Enterprises
Interstellar Alcohol Conglomerate
|
Posted - 2006.10.11 17:21:00 -
[73]
I think pershphanie hit the nail on the head with the statements he/she made. That would help ten-fold
I also know there is a chip being made that suppose to revolutionize gaming as we know it.
I know with added physics and image rendering, technology sorta hit a peak on really what they can do. Like I believe the most advance gaming engines can only run so many entities on screen that allows each piece to have a "physics equations" added to each when certain action coding happen to it. The developed chip is suppost to run all rendering and physics in it so that not just the processors and video cards are rendering all the codes and processes. If this chip does happen to be finally built, and marketed, then games and massive games like eve should balance out. Course you'll have to upgrade your motherboard to fit this chip in lol
Regardless of the fact, I did notice lag started getting worse when they updated the resolution to the ship models and added ships (though the added ships shouldnt be a problem)
And with Kali, the model resolutions are said to be INCREASED again!
Think eve is broke now, think what this might do
Not only that there is only high resolution settings for the eve-client.
If CCP is reading this, I will ALWAYS! ALWAYS!!! sacrifice eye-candy graphics for a extremely fun and playable game. I'd do that anyday
|
Rikeka
Amarr
Eye of God
Axiom Empire
|
Posted - 2006.10.11 17:26:00 -
[74]
If I recall correctly, though. Back then it WAS possible for more numbers on the system before a node crash, as there were some serious fleet battles there (for what I was told)
|
Yodaron Ballsithor
Gallente
Black Avatar
Firmus Ixion
|
Posted - 2006.10.11 18:05:00 -
[75]
Interesting topic that I know very little of.
What I do know is that the 2 IAC posters that say things like "No, No, No" need to read, comprehend and then respond. This topic is not about IAC or any one particular alliance. It is about the struggle of fleet warfare - heck, the struggle for 20 v. 20 according to some.
Keep the crap out of this thread and maybe we will see again another intelligent discussion like appears to be happening here.
If you desire to flame someone, start your own thread or pick one of the hundreds that are riddled with them.
I, for one, and learning from reading the well put together and cogent thoughts and suggestions that are in this thread. Let us keep it that way.
 |
travelingsales
Ars Caelestis
Ascendant Frontier
|
Posted - 2006.10.11 18:15:00 -
[76]
Edited by: travelingsales on 11/10/2006 18:16:26
Originally by: Nez Perces
Originally by: travelingsales
To be honest the first couple of times I remember hearing about large numbers of node deaths were during the Goons vs. D2 war in the north. How long ago was that? I seem to recall it was right around when Dragon came out, which would fit with the general theory that something went wrong with Dragon.
Can anyone with a better memory than me confirm the timeline?
Actually the dragon code was introduced during the Goons vs D2 war (could be wrong .. but dont think so)... I do remember some D2 members hoping that the dragon code would solve their problems. The goon vs D2 war was plagued by problems from the start and there was no dragon code when it kicked off... ( i think..)
The main reason for the horrific problems with Goonfleet and D2 was the massive numbers GF were fielding.
Would be nice to hear more from D2 or GF about this.
I still think raw numbers are not the whole story though. To bring up the 9cg battle once again, at one point there were over 400 players fighting in local there (it was confirmed that at one point there were more people in 9cg than in Jita). The battle was pre-Dragon, and while it was not POS centric (there was no spamming) there were at least 6 or 7 POS in the system and Dreads were deployed (there was actually some dread-on-dread combat, it was a good fight all around). As far as I can recall the node never crashed, or at least it didn't crash during any of the times when I was there (there are players from about 8 different alliances who may be able to add more detail to this )
I was not at any of the GF/D2 battles but from what I heard they never exceeded the 400 in local that occured in 9cg. It would be nice to have some other data points to work from as far as REALLY large fleet battles, but sadly all the other ones I have been to with 300+ in local were post-Dragon, and all the other pre-Dragon ones I was at were around 100-120 per side, so only about 200-230 in local. However those were all fine, and judging from my current experience of the ASCN/BoB war as well as reading about the MC's recent events, it seems even 100v100 will often crash the server these days.
|
Murukan
Minmatar
The Priory
Interstellar Alcohol Conglomerate
|
Posted - 2006.10.11 20:14:00 -
[77]
Edited by: Murukan on 11/10/2006 20:14:24
Originally by: Yodaron Ballsithor
Interesting topic that I know very little of.
What I do know is that the 2 IAC posters that say things like "No, No, No" need to read, comprehend and then respond. This topic is not about IAC or any one particular alliance. It is about the struggle of fleet warfare - heck, the struggle for 20 v. 20 according to some.
Keep the crap out of this thread and maybe we will see again another intelligent discussion like appears to be happening here.
If you desire to flame someone, start your own thread or pick one of the hundreds that are riddled with them.
I, for one, and learning from reading the well put together and cogent thoughts and suggestions that are in this thread. Let us keep it that way.
Uh i very much read and comprehended the original poster. If you read my post you would see that i talked generically about a defender. To ask that anyone defending their space hold back so the invader can have fun is stupid. So don't try and act like i'm some idiot and you're above me because i fully understood the original post and frankly it's crap.
In rust we trust!!! |
Yucali
Candy Hearts
|
Posted - 2006.10.11 21:53:00 -
[78]
I agree completely with the OP, tbqh BoB, MC and FIX shouldn't have to go through this messy business of fighting for regions
IAC and ASCN: plz line up in rows against the wall and wait quietly till its your turn to be dispatched
|
DB Preacher
Reikoku
Band of Brothers
|
Posted - 2006.10.11 22:13:00 -
[79]
Game is broken not the alliances.
We'll keep pushing Eve to it's limits and I can't wait for the first 1000 enemies vs 200 BoB.
That will pwn.
dbp
Caldari Alliance PVP Championship Winner
Current RKK Ranking: (PSCAL6) Proficient Short Tanto
|
Mistress Suffering
Einherjar Rising
Dusk and Dawn
|
Posted - 2006.10.11 22:20:00 -
[80]
My impression is that it got notably worse after Dragon. Whereas prior to that time, you could run a decent large gang action (20 v 20) without undue lag, after Dragon such engagements became utter sludge. Yes, hundreds of ships on both sides has been slow for quite a while, but it is my impression that as of Dragon it got worse.
Plus Dragon brought us the utter abomination that is jump queues. These are a total joke and should not even begin to exist.
Telling players to solve these issues by "not bringing so many guys" is not the answer. All that does is ask the honorable players to get reamed by the dishonorable. "Ok, we'll only bring 20 guys" - Enemy shows up with 30, or 3 groups of 20, etc... Forget that. If the game encourages team play with obstacles (POS) that can only be reasonably down in large numbers, you will see large numbers. The more important the conflict, the greater these numbers will be.
CCP can address the problem in a number of ways but they made need to open their thinking. They may need to chunk space smaller and improve their transition logic. They may need to split the server into multiple shards so they can get fewer players per server machine. They may need to reduce the information being sent to each player's machine and do more prediction on the client. They may need to look at the POS code, or cause X number of players in a system to cause unpiloted objects (cans, shuttles, etc..) to be deleted. There are a great many options, but these options need to be explored, because the game right now is broken and annoying because of it.
|
Erichk Knaar
Caldari
Tactical Enterprise Group LTD
The Imperial Order
|
Posted - 2006.10.12 03:42:00 -
[81]
The only thing I can add to the excellent posts Semkhet made is this:
1) Python, I believe, can be capable of the scale CCP is trying to achieve. I don't think everything should be re-written, just the performance critical stuff. Pythons interfaces to native code are excellent and can be leveraged to keep the flexiblity, but with the hard stuff done in the most efficient way possible.
2) Dragon is when every single string in Eve turned into unicode. 2 bytes vs. 7 bits. For the unaware, this means that almost evey piece of text in Eve is now double the size when being passed around the system. This is where I believe the root of the problem lies.
|
Yazoul Samaiel
Caldari
Black Nova Corp
Band of Brothers
|
Posted - 2006.10.12 04:00:00 -
[82]
Imo i was thinking about the issue from a differen direction , as a programmer when you look at the whole mass jump in system or login and the lag starts to spike you realy have to ask what is the common factor we dealing with here . BOB VS Toxin/Celes was not lag free , at least not for me , it was a lag festival and the system was crap even with 22 ppl in local  I think that BMs are the major criminal here in this lag and node crash probelm.
Here is some intresting facts:
-Mass copying BMs have been noted as the most effective way to crash node and system , not sure if it is marked as exploit or not but i rememebr it was done in jita as a mass protest and crashed the system repeatedly , also some lamers use it from tiem to tiem to crash battels.
-Fact , copying bms now even when you are docked laggs the client and lags the system so it is both a client and server side issue.
-I dont know how BMs are dealt in the coe wise , using arrays or hash tabels etc but here is one important fact to consider., In every major fleet battle or medium sized engagemnt spehcialy during the jumpin , u have like 20 ppl each one of them carrying at least 5 k bms (Pretty sure some ppl have max 10k )they all jump into system and server has to search categorise these bms and see if they match current system or not !!! Just imagien the CPU load and search processes and arguments going back and forth so when u load up ur screen u only see the BMs in that system !!!
-IIRC BM are an exploit to the original system and they werent intended to be , so beats me why they are still used ???
-IMO BMS should be removed and warp at 1.5 km to object option to be added , as for ppl who whine about empire wars saying they wont be possible , well hello bms are sold legit on escrow and ppl have BMs and can run in empire as they want just coz 1 or 2 war targets dont have the bms validates ur argument.
-Also i have no idea about load balancing methods CCP uses but why not auto allocate node strengh etc to systems that are having mass activity judging from CPU utilization? I think this has been started recently sicne Dragon patch spechialy that system with no ppl go offline thus saving node strengh , u can see it if u logg off in a rmeote system and try to log back later on u will see a msg saying system being loaded , this was a brilliant idea .
I realy suggest that CCP attempts a Stress test with 200 VS 200 ppl with no BMs just for factor eliminating coz i am willing to bet my top dollar that BMs are the most guilty party here .
"There is no such thing as innocence , only different degrees of guilt"
|
James Potkukelkka
BGG
Freelancer Alliance
|
Posted - 2006.10.12 05:39:00 -
[83]
Originally by: Vegas
Eve's broke.
But no lag on collecting money for my subs
Go figure
Oh how true 
But! remember guys, you can vote with your wallet. That is the only way that will be heard: Set an extra long skill on and let sub run out. Come back when its playable again.
|
Sir Vivor
|
Posted - 2006.10.12 08:17:00 -
[84]
A possible fix could be to set grids into a sort of throttled "the matrix" mode when there are too many ships. Though this might feel "laggish," the system will not crash due to data saturation. A factor of two, for example, would mean one second in Eve time is equal to two seconds of real time. This effectively cuts the amount of data being transferred between eve and clients by half (relative to time).
This would, however, gives pilots an advantage because they now have twice as long (with factor of two) to make a decision.
A factor of two might be un-necessary. The throttling could be dynamic based on a saturation level (ship count, etc.) and might come out to something like 1.1 or 1.2 which wouldn't be as "noticeable" to the user. It is a lot better then hitting that saturation level where the system can't keep up with the data stream and starts to degrade exponentially...
Of course, Eve could already be doing this...
Just a though..
PS - But wow would it be interesting to deal with "time inconsistencies" between grids.
|
kDaser
Caldari
FATAL REVELATIONS
|
Posted - 2006.10.12 08:32:00 -
[85]
Yazoul Samaiel makes a very valid point with bookmarks. I agree that they are a big part of the issue. I, having no programing knowledge at all want to know if dragon being intended for Kali is causing the the other half of this issue and if it will resolve the issue by updating to Kali?
|
| |
|
| Pages: 1 2 [3] :: one page |
| First page | Previous page | Next page | Last page |