Pages: [1] :: one page |
|
Author |
Thread Statistics | Show CCP posts - 0 post(s) |
TheBlueMonkey
Gallente Priory Of The Lemon Atlas Alliance
|
Posted - 2010.07.12 12:31:00 -
[1]
So it seems that todays flavour of the week is *****ing about how crap eve is and how each and everyone of us could write a better game and that's why we all run a multinational company with many employees... no? just me... no.. me either.
The Point I was musing the other day after being shown some rather cool stuff that virtual servers could do and was wondering what people thought (obviously it's a **** idea but I'd like to know why and how)
The presentation I saw had a number of virtual servers running on a number of blades. These virtual servers were then pushed between the blades, without interuption to service, as the loads changed. At one point all the servers were running on less than 10% cpu and got pushed onto one blade and the blades that had nothing on them shut down to conserve heat and power.
What if eve were written that each system was held by a single virtual server, then all of these virtual servers were housed on a large(ok, giant) blade enclosure. This would make "reinforcing nodes" redundant as when fleets moved between systems, the systems(or virtual servers) would be moved to a blade with minimal cpu usage. You'd obviously need some kind of chunky fibre backbone in place between the servers too.
But ultimately it'd reduce the "zomfgblackscreen5minwaitboom" that alot of 0.0 dwellers get when a 400 man fleet turns up.
Although Jita should probably just have it's own server anyhow.
I'm guessing that this idea would require an epic rewrite of the eve code but from the general grumblings it sounds like we're heading that way anyhow. I mean, when eve was first written, what population was it supposed to support? --
Nothing is worthless, you may have gotten it for free but it still has an inherent value
|
omgevenmoarfreemoniez
|
Posted - 2010.07.12 12:32:00 -
[2]
LOL EVE is fine, stop whining. If you don't like it, don't play it.
|
rowbin hod
Cloak and Daggers Honourable Templum of Alcedonia
|
Posted - 2010.07.12 12:37:00 -
[3]
Interesting.
I've seen live VM migration work first hand (VMWare) so yeah, in theory it could work. The beauty is that nothing requires a recode - the VM just looks like a machine to the OS, and it's the hypervisor + associated software that takes care of shifting the VMs around, completely independantly of the guest OS, which knows nothing about it.
However, I would expect that in a game like Eve there would be a performance hit during live migration, and I don't know how long live migration would take, so it may well create more problems.
Still, nice idea - VMs have come a long way in recent years.
|
Dana Jin
|
Posted - 2010.07.12 12:43:00 -
[4]
I'm sure CCP employees know very well what is going on in the virtualization sector...particularly for servers...hell I'm no game DEV or even in the IT industry and I know where it's headed. The devs have hinted at wanting to be able to push nodes around to unused resources for a while now. I'm sure what you suggest has been or is being considered. But obviously as you said it would take a rewrite of the code. From my understanding EVE was originally written in PYTHON and a good majority of it still is. I believe they are slowly converting what they can to a more friendly language and looking into virtualization.
|
Ressiv
Cooperative Freelance Navigators Association
|
Posted - 2010.07.12 12:46:00 -
[5]
Originally by: TheBlueMonkey XXth reitteration on VM's
.. you are aware of the search feature ? ========================== Nothing is true, everything is permitted. ========================== |
TheBlueMonkey
Gallente Priory Of The Lemon Atlas Alliance
|
Posted - 2010.07.12 12:55:00 -
[6]
Originally by: Ressiv
Originally by: TheBlueMonkey XXth reitteration on VM's
.. you are aware of the search feature ?
Seeemd like a fun band waggon to jump on and I was somewhat hoping for a dev to wander in and mention something in passing.
Rather than watching another day go by where everyone just *****es about eve being **** and laggy.
It's an interesting topic too, well, I find it interesting and looks like it offers a decent solution. --
Nothing is worthless, you may have gotten it for free but it still has an inherent value
|
Musical Fist
Gallente NAP Coalition HYDRA RELOADED
|
Posted - 2010.07.12 12:56:00 -
[7]
ITT The Irony burns my eyes
MEGA HUGE WTF NAPs + Blobbing = Node Crashes
Node Crashes almost never (that doesnt mean they havent) affect small gang warfare, I know there are several 100 man ops etc that have been affected by Node Crashes but all in all it is quite stable.
So lets look at the formula again
MEGA HUGE WTF NAPs + Blobbing = Node Crashes
and again
MEGA HUGE WTF NAPs + Blobbing = Node Crashes
and one last time
MEGA HUGE WTF NAPs + Blobbing = Node Crashes
Now I know people will get confused and compare PL / Atlas to Afghanistan / Chernobyl but this is a game it isnt RL so stop complaining about lagg and start resetting.
I feel 0.0 mega huge WTF powerblocks enjoy node crashes soo much thats the reason they stay in 50k+ NAP trains.
Besides if you cant win a 400 man vs 90 encounter I think its time to pack up and go home hahaha fixing lag will just allow you to unskilfully hotdrop anyway.
Say no to blobs yes to skill! --
Recruiting now open!! |
Ressiv
Cooperative Freelance Navigators Association
|
Posted - 2010.07.12 13:08:00 -
[8]
Originally by: TheBlueMonkey I was somewhat hoping for a dev to wander in and mention something in passing.
Read up on the Dev Blogs, or just look for the posts around the last few patch days. Devs did reply there, with the reasoning why 3rd party VM sollutions will not work for EVE.
Google: 'eve online devblog server virtualisation' and get this ========================== Nothing is true, everything is permitted. ========================== |
Diosas
|
Posted - 2010.07.12 13:19:00 -
[9]
Well i do have alot of experienced with VM,seeing im a CIO/IT Director. And VM really is the way to go now, it used to be slow and well a pain but has improved immensley. In fact alot of large scale environment use VM. The flexibility and ease of backing up, cloning, setting up is a million times easier, and almost fault free. Only thing is is have meaty host servers but hey thats easy now. So eve on VM most definately..
|
Tres Farmer
Gallente Federation Intelligence Service
|
Posted - 2010.07.12 13:31:00 -
[10]
Edited by: Tres Farmer on 12/07/2010 13:32:11
Originally by: Diosas Well i do have alot of experienced with VM,seeing im a CIO/IT Director. And VM really is the way to go now, it used to be slow and well a pain but has improved immensley. In fact alot of large scale environment use VM. The flexibility and ease of backing up, cloning, setting up is a million times easier, and almost fault free. Only thing is is have meaty host servers but hey thats easy now. So eve on VM most definately..
Didn't the last link sink in? In the post above yours?
EVE already is as virtual as possible to actual standards/possibilities of CCP and the way they'd choose for the backbone of the system. The smallest block of code comprises 1 solar system on one core and can run around 1,400 people. Not less space, not more people.
|
|
Nareg Maxence
Gallente
|
Posted - 2010.07.12 13:57:00 -
[11]
Devs said when this was last brought up, that it wouldn't work for the EVE server.
|
TheBlueMonkey
Gallente Priory Of The Lemon Atlas Alliance
|
Posted - 2010.07.12 13:57:00 -
[12]
Originally by: Tres Farmer run around 1,400 people.
I'd like to see that in 0.0
well... with 700 a side... not 1300 vs 100 --
Nothing is worthless, you may have gotten it for free but it still has an inherent value
|
Musical Fist
Gallente NAP Coalition HYDRA RELOADED
|
Posted - 2010.07.12 14:01:00 -
[13]
Edited by: Musical Fist on 12/07/2010 14:03:29
Originally by: TheBlueMonkey
I'd like to see that in 0.0
well... with 700 a side... not 1300 vs 100
Also loling at the fact that you actually want to see equal numbers in a fight, face it you can remove lagg completely from the game and all you will get is epic 10k vs 20 battles think its time we look at the real problem here because lag well something tells me that isnt getting fixed for a reason --
Recruiting now open!! |
Diosas
|
Posted - 2010.07.12 14:08:00 -
[14]
Originally by: Tres Farmer Edited by: Tres Farmer on 12/07/2010 14:02:24
Originally by: Diosas Well i do have alot of experienced with VM,seeing im a CIO/IT Director. And VM really is the way to go now, it used to be slow and well a pain but has improved immensley. In fact alot of large scale environment use VM. The flexibility and ease of backing up, cloning, setting up is a million times easier, and almost fault free. Only thing is is have meaty host servers but hey thats easy now. So eve on VM most definately..
EVE already is as virtual as possible to actual standards/possibilities of CCP and the way they did choose for the backbone of the system wayyyyy back: Second Genesis was released 2003.. until that time 2..3..4..5GHz+ processors where visioned as the future.. 2005 the first dualcores show up and parallel computing became the way to go.
The smallest block of code comprises 1 solar system on one core and can run around 1,400 people. Not less space, not more people.
Personally I would bet 100M isk that in 2 years time we'll experience a grid per core and maybe some very fancy/mad idea how to shrink that even smaller..
I think you have this confused mate, this is about Virtualisation technology like VMWare that use virtual servers not native 'actual physical boxes' platform. I think you thinking of this from a Virtual Gaming World scenario.
|
Barakkus
Caelestis Iudicium
|
Posted - 2010.07.12 14:13:00 -
[15]
Originally by: Dana Jin I'm sure CCP employees know very well what is going on in the virtualization sector...particularly for servers...hell I'm no game DEV or even in the IT industry and I know where it's headed. The devs have hinted at wanting to be able to push nodes around to unused resources for a while now. I'm sure what you suggest has been or is being considered. But obviously as you said it would take a rewrite of the code. From my understanding EVE was originally written in PYTHON and a good majority of it still is. I believe they are slowly converting what they can to a more friendly language and looking into virtualization.
You're limited by the hardware still, and you will take a performance hit in addition to it not being exactly reliable either.
VMs are fine if you aren't using up all of a machine's resources, and you can deal with the machine running the VMs going down and taking everything with it.
Originally by: CCP Dropbear
rofl
edit: ah crap, dev account. Oh well, official rofl at you sir.
|
Ressiv
Cooperative Freelance Navigators Association
|
Posted - 2010.07.12 14:20:00 -
[16]
Originally by: Diosas Well i do have alot of experienced with VM,seeing im a CIO/IT Director. And VM really is the way to go now, it used to be slow and well a pain but has improved immensley. In fact alot of large scale environment use VM. The flexibility and ease of backing up, cloning, setting up is a million times easier, and almost fault free. Only thing is is have meaty host servers but hey thats easy now. So eve on VM most definately..
Without trying to insult you, generaly knowledge resides on the workfloor, not above. More often then not, the reason for things going haywire are directors who think on the technical side.
Your response showed you didnt bother to look into reply's made in topics that can be found in that google search, and provide not one single argument in favor of migration to a different VM system then is being used now.
So EVE on VM, already the case ... just not any off-the-shelf solution.
To provide you with a small hint: The hypervisor creates unneeded overhead compared to the current system. ========================== Nothing is true, everything is permitted. ========================== |
iCrusher
Gallente Heresy Inc.
|
Posted - 2010.07.12 14:26:00 -
[17]
Kind of ironic that the Devs won't run a VM solution because of the increased overhead.
Yet they have Windows servers ...
|
Durov
|
Posted - 2010.07.12 14:36:00 -
[18]
Edited by: Durov on 12/07/2010 14:38:34 Edited by: Durov on 12/07/2010 14:36:59
Originally by: TheBlueMonkey The presentation I saw
This is where you went wrong.
Virtualisation is a great solution when you have a large number of small servers that have very low resource requirements e.g. mail servers. This lets you run them all on a single physical server, saving you money. Tranquility has entirely different needs - it needs larger physical servers, so it can hold more pilots at any given time. VMware cannot allow a virtual machine to grow larger than any single physical host, so there is no benefit to be gained.
Read up on the differences between Horizontal scaling and Vertical scaling.
|
second toLate
|
Posted - 2010.07.12 15:19:00 -
[19]
I am no expert but wouldn't the addition of the virtual layer not cover the benefit of the virtualisation like the OP said ?
????
|
Ressiv
Cooperative Freelance Navigators Association
|
Posted - 2010.07.12 15:24:00 -
[20]
Originally by: second toLate I am no expert but wouldn't the addition of the virtual layer not cover the benefit of the virtualisation like the OP said ?
????
No. ANY type of governing layer, to handle the VM switchovers, will reduce max cappacity of the servers, and thus render any possible gain useless.
On top of that, there are issues moving sessions around on VM's, according to CCP. CBB to look up the quote, dig through the google link I provided earlyer and you can find it somewhere in there.
========================== Nothing is true, everything is permitted. ========================== |
|
Taedrin
Gallente Xovoni Directorate
|
Posted - 2010.07.12 15:27:00 -
[21]
First off, I'm no dev. I'm just a student. However, the issue that I see is that virtualization doesn't magically turn multiple blades into one super fast server. The virtual machine is comprised of multiple blades, true. However, the virtual machine is going to look at each blade as if it were a separate CPU. We will STILL have the problem that CCP currently has: one solar system can be allocated on at most ONE CPU core.
What virtualization might do is enable dynamic load balancing (noticing that a fleet battle is occuring without player/dev input and automagically moving that solarsystem over to a reinforced node). ----------
Originally by: Dr Fighter "how do you know when youve had a repro accident"
Theres modules missing and morphite in your mineral pile.
|
TheBlueMonkey
Gallente Priory Of The Lemon Atlas Alliance
|
Posted - 2010.07.12 15:58:00 -
[22]
Originally by: Taedrin Edited by: Taedrin on 12/07/2010 15:29:13 First off, I'm no dev. I'm just a student. However, the issue that I see is that virtualization doesn't magically turn multiple blades into one super fast server. The virtual machine is comprised of multiple blades, true. However, the virtual machine is going to look at each blade as if it were a separate CPU. We will STILL have the problem that CCP currently has: one solar system can be allocated on at most ONE CPU core.
What virtualization might do is enable dynamic load balancing (noticing that a fleet battle is occuring without player/dev input and automagically moving that solarsystem over to a reinforced node).
EDIT: and as has been already mentioned, adding virtualization adds an additional layer which consumes resources, which will reduce the maximum performance. This, however, does NOT mean that it isn't worth it. Dynamic Load Balancing has great opportunities.
Indeed, the limit is the fastest blade you have in your cluster but then you could keep multiple "quiet systems" on a sigle blade and have each of them moved to a blade that isn't being used as CPU starts to get used more and more.
From the things I've seen the switching virtual servers between physical hardware is mighty impressive. The example I was shown was a number of files being copied from VC1 to VC2, then, mid copy they pulled the power on the physical blade that VC1 was hosted on. The virtual machine software moved VC1 to a new blade without interupting the copy.
Also, the VC's would be moved around based on the first few members of fleet jumping into system so if there was any additional overhead from moving VCs I'd expect it there which has to be preferable to moving a fleet into a system and then popping people that lag out as they jump in.
Oh and also if CCP want us to fight in smaller fleets, they should probably stop advertising huge fleet battles as one of the selling points. --
Nothing is worthless, you may have gotten it for free but it still has an inherent value
|
Nareg Maxence
Gallente
|
Posted - 2010.07.12 16:14:00 -
[23]
Here is CCP Yokai's answer.
|
Jaina Kort
|
Posted - 2010.07.12 16:24:00 -
[24]
Originally by: Diosas Well i do have alot of experienced with VM,seeing im a CIO/IT Director. And VM really is the way to go now, it used to be slow and well a pain but has improved immensley. In fact alot of large scale environment use VM. The flexibility and ease of backing up, cloning, setting up is a million times easier, and almost fault free. Only thing is is have meaty host servers but hey thats easy now. So eve on VM most definately..
I am also an IT Director and have worked with these types of systems and high availability systems, but I respectfully have to disagree with you on this. VM is a great solution for certain systems. But, from reading old dev blogs and having been to previous fanfests where they were showing the cluster now and their hopes for supercomputing, I would have to disagree with you there.
First, you would have to know what you wanted to virtualize. If you were to virtualize the whole of eve, it would not be just throwing the whole cluster onto a bunch of vm'd servers and tell it to have fun. There is routing, services running on nodes, system like jita itself that need all the processing it can get.
Ok, I have listed just a few things and I am sure there are a lot more that they have not shared with us. But, the big thing and this is huge.... Python and the global interpreter lock. You cannot trick python into thinking that a vmware cluster is only 1 core on 1 CPU. That is not the way threading works.
As ,CCP has stated prior the problem with Jita is that although you can have a 4 or more core CPU, they can only use 1 of the cores and not all 4 of them.
If we looked at how ccp is running the cluster, having everything on their own boxes now, we would have to understand the services that are running. Ie: do they all run with the same service name or their own unique name?
I would say that if they were not to sit down and either make their own version of python without the lock or rewrite in another language that did not have the lock, then they would do much better by parallelization of the system. This meaning that they take part of the code and tell it to run on different cores or processors in the box.
Any way they go, if they want to up the limits of system such as jita, the ways above are the ways I would think they could do it. There is no magic solution on a very complicated issue and I am sure there are a lot of things we don't even understand that are in the code, services, and even transport. But, VM I could not see as doable.
|
H3llHound
Blacksteel Mining and Manufacturing DRACONIAN COVENANT
|
Posted - 2010.07.12 16:24:00 -
[25]
What I would like to see is the ability that TQ automatically shifts solar systems with more than average activity to a stronger node with out the need of a DT. This could make small fleet battles and gang fights(up to 200pilots) that arent planned 48h before not end up in a lagfest like now.
Ofc this wouldnt improve huge fleet fights blobs. Dont blob!! Recruiting │My 3rd Party Service |
Nareg Maxence
Gallente
|
Posted - 2010.07.12 16:28:00 -
[26]
Originally by: H3llHound What I would like to see is the ability that TQ automatically shifts solar systems with more than average activity to a stronger node with out the need of a DT. This could make small fleet battles and gang fights(up to 200pilots) that arent planned 48h before not end up in a lagfest like now.
Ofc this wouldnt improve huge fleet fights blobs. Dont blob!!
They are working on this. Try reading CCP Yokai's replies in the thread linked above.
|
Norian Lonark
Gallente Black Thorne Corporation Black Thorne Alliance
|
Posted - 2010.07.12 16:34:00 -
[27]
I'm glad CCP have access to these forums so that they can get all the advice they need on how they should be developing their infrastructure and technologies.. what would they do without them
|
Jaina Kort
|
Posted - 2010.07.12 16:50:00 -
[28]
Edited by: Jaina Kort on 12/07/2010 16:51:47
Originally by: Norian Lonark I'm glad CCP have access to these forums so that they can get all the advice they need on how they should be developing their infrastructure and technologies.. what would they do without them
Ha ha ha very true. I am sure they stay up at night clicking on these posts to solve their challenges. Ha ha.
But, I think that this is the best video http://www.eveonline.com/download/videos/Default.asp?type=7&order=2 that they have out on the cluster and what they were thinking at least in 2008. Look for the video named TQ Servers: Making a mountain out of a molehill ... Or something like that. Too lazy to look it up, but should be easy to find.
|
|
|
|
Pages: [1] :: one page |
First page | Previous page | Next page | Last page |