open All Channels
seplocked EVE Information Portal
blankseplocked New Dev Blog by Mindstar: What's this about Nodes?
 
This thread is older than 90 days and has been locked due to inactivity.


 
Pages: 1 2 [3] 4

Author Topic

Deva Blackfire
Viziam
Posted - 2008.10.07 11:14:00 - [61]
 

Im quite happy with results but i have one thingy:
why most clients blackscreen on system load while jumping in/logging on? Which part of server architecture is responsible for it? And finally - is it possible to see fix in forseable future? ;p

Moon Kitten
GoonWaffe
Posted - 2008.10.07 11:49:00 - [62]
 

Originally by: Mioelnir
Originally by: Necronomicon
None of this is going to stop CCP employee xyz contacting Alliance Leader abc on MSN/IRC and telling them though, is it?
And nothing is going to stop Alliance Leader abc from replying: "Yeah, yeah, I know already."

There is no secrecy in eve.

I guess that makes it ok for CCP employees to cheatRolling Eyes

Mr Horizontal
Gallente
Posted - 2008.10.07 11:55:00 - [63]
 

Can I ask how many of these new Wolfdale SOL blades will be made available? Have you done any analysis of the general amount of fleet fights to get an idea of worst case and numbers of fights that go on across the galaxy?

I only ask because if there are say 16 different systems with major fleet fights around EVE at a peak time, then you're gonna need at least 8 blades dedicated (RAM dependent).

Also, while the limit is much higher, there's still a limit on the number of pilots the server can handle - if we start notifying CCP of plans, and shove 1500 people into a system from one side, then that also only leaves the other side with 200 people, thus making the fight a one-sided massacre won by whoever clears the jump queue first.

Deva Blackfire
Viziam
Posted - 2008.10.07 11:59:00 - [64]
 

@ Moon Kitten - i know you live in some state of paranoia about devhax but you have 2 choices: either to tell CCP to which systems to reinforce or dont. If you dont, you will get node crash and reinforced node just after, and then you will play "who can log on 1st" game. If you like the 2nd option - be my guest.

Moon Kitten
GoonWaffe
Posted - 2008.10.07 12:25:00 - [65]
 

Originally by: Deva Blackfire
@ Moon Kitten - i know you live in some state of paranoia about devhax but you have 2 choices: either to tell CCP to which systems to reinforce or dont. If you dont, you will get node crash and reinforced node just after, and then you will play "who can log on 1st" game. If you like the 2nd option - be my guest.

I would not call it paranoia. The combination of CCP's history and their policy of letting their employees play Eve and the widespread practise of meta-gaming in Eve makes it incredibly naive to not ask about who at CCP has access to information about future fleet movements.

There are more than two outcomes. Depending on how large your fleet is and your opponents response it would not necessarily result in lag or node crashes.

Eraggan Sadarr
Comply Or Die
Posted - 2008.10.07 12:47:00 - [66]
 

Hey Mindstar. A really nice read. The tech level of the blog was also very good and digestible for most i would think. And it's always interesting to read about the hardware/software that makes this fantastic universe possible Smile

Regards
Eraggan Sadarr

Frug
Omega Wing
Snatch Victory
Posted - 2008.10.07 13:14:00 - [67]
 

A winning blog, from the meme in the title down to the content. pwnage.

Nicholas Barker
Deez Nuts.
Posted - 2008.10.07 14:32:00 - [68]
 

well, we'll never be able to get away with lagging the systems out now, i guess we'll just have to sit titans on the entry gates to the target system i guess.

Apple Boy
Gallente
Tri-gun
Auctorita Alliance
Posted - 2008.10.07 14:38:00 - [69]
 

agreed, the tech blogs are the best blogs Smile

As for everyone calling spies in CCP, etc. you have two choices:
1) don't tell them, lag out
2) tell them, trust them, and have a higher chance of having a fun fleet fight.
they didn't say anything about it being mandatory to tell them.

What I find amusing is everyone is complaining that they the GMs will tell their enemies that they're going to attack a place. If you're asking for a node to be reinforced, it's usually because someone has reinforced a POS and you're expecting that your enemy will show up. I hardly see how that intel will make any difference. If you want a fun fight, tell them. If you want to try and do some stealthy operation, chances are you won't be getting a huge fleet fight and won't need a reinforced node anyway. Rolling Eyes

Chaos Incarnate
Faceless Logistics
Posted - 2008.10.07 14:41:00 - [70]
 

Originally by: Moon Kitten
I would not call it paranoia. The combination of CCP's history and their policy of letting their employees play Eve and the widespread practise of meta-gaming in Eve makes it incredibly naive to not ask about who at CCP has access to information about future fleet movements.



Uh, I'm betting that a good number of those CCP employees could get access to your evemail, ingame chat logs, ship locations, assets, and other important info that could be leaked to various alliances. Freaking out that they'd have another potential point of access to the same info is pretty moot - if they wanted to leak your sekrit intel, they'd be doing so already.

Jallem Sims
Minmatar
Native Freshfood
Posted - 2008.10.07 15:48:00 - [71]
 

"It is worth noting that the hardware we are beginning to purchase now is the hardware that will see us all the way into the HPC era"


HPC means?

thanks, and top dev blog!

ShadowMaster
Gallente
Posted - 2008.10.07 16:05:00 - [72]
 

Edited by: ShadowMaster on 07/10/2008 16:05:42
Google:
define:hpc
Quote:
High Performance Computing. Generally denotes computing that approaches and/or passes the TeraFLOPS performance range.


Enjoy.

Edit: Also, very nice dev blog. I really enjoy the recent flood of dev blogs. Keep up the good work.

Gavin Darklighter
Ministry of War
Posted - 2008.10.07 16:31:00 - [73]
 

ubernode ftw?

Tanathryn
Caldari
The 5th Freedom
Wildly Inappropriate.
Posted - 2008.10.07 17:15:00 - [74]
 

What is the likelihood, in the future, of being able to remap systems on the fly to dedicated sol servers? The cluster sees a rapidly increasing population in a 0.0 system and is able to move the system to one of several standby servers set up just to support these situations without requiring a downtime to reconfigure the nodes. That would be an amazing capability.

Synapse Archae
Amarr
Viziam
Posted - 2008.10.07 18:21:00 - [75]
 

Tonight, let us send a message to that lag. Let us shake these forums of code and database from hard drive to PHP script!

Let us remind them,

THIS IS EVE ONLINE, AND WE ARE NOT AFRAID!

DrAtomic
Atomic Heroes
Chain of Chaos
Posted - 2008.10.07 18:41:00 - [76]
 

Awesome blog! Want moooooooooooar! Funny thing I received my EON today and read the Mindstar profile which had me thinking finally a sysadmin getting some attention, silly bug coders always getting all the attention, we can kill applications a lot better then devs!

Aaaaanyway, did your wife move to Iceland as well?

DrAtomic
Atomic Heroes
Chain of Chaos
Posted - 2008.10.07 18:50:00 - [77]
 

Edited by: DrAtomic on 07/10/2008 18:50:46
Originally by: ShadowMaster
Edited by: ShadowMaster on 07/10/2008 16:05:42
Google:
define:hpc
Quote:
High Performance Computing. Generally denotes computing that approaches and/or passes the TeraFLOPS performance range.


Enjoy.

Edit: Also, very nice dev blog. I really enjoy the recent flood of dev blogs. Keep up the good work.


In case of EVE they mean recoding the server code to be able to process parallel, or in other words make the servercode multicore aware i.e. multithreaded.

Also of note Intell is working on a technology that makes single thread apps multi threaded without code intervention: http://apcmag.com/we_can_transform_single_thread_to_multithread_intel.htm I dont know how far they are with that though and/if the devs have even considered that solution.

edit: wtb autolink feature on the forums

CCP Mindstar

Posted - 2008.10.07 19:10:00 - [78]
 

Originally by: DrAtomic
Aaaaanyway, did your wife move to Iceland as well?


While we did "meet" in a bar, I am not entirely sure I remember getting married.... I'll have to double check on that one I think.. Where do you get your information?? Very Happy

adriaans
Amarr
Ankaa.
Nair Al-Zaurak
Posted - 2008.10.07 20:46:00 - [79]
 

sweet stuff Very Happy


Ethaet
Gallente
Aliastra
Posted - 2008.10.07 21:35:00 - [80]
 

Nice blog, very interesting Smile

Also,
My Raven was equipped with the following:

HIGH
06 x Cruise Missile Launcher I
01 x SMALL TRACTOR BEAM 1
01 x SALVAGER I

MEDIUM
04 x LARGE SHIELD EXTENDERS
01 x 'HYPHNOS' ECM
01 x MEDIUM SHIELD BOOSTER

LOW
01 x EMERGENCY DAMAGE CONTROL
01 x ARMOR KINETIC HARDENER I
01 x ARMOR THREMIC HARDENER I
02 x WARP CORE STABILIZER I

DRONES
02 x WARRIOR I DRONES
03 x HAMMERHEAD I DRONES

UPGRADES
01 x ROCKET FUEL CACHE PARTINTION I
01 x BAY LOADING ACCELERATOR I

Pacala
Gallente
Evolution
IT Alliance
Posted - 2008.10.07 22:04:00 - [81]
 

Thank you for the detailed update. As you've matured as a company, and EVE has come a long way as a product, these blogs have become a fantastic window into your efforts.

Some comments on overall strategy though. It's clear that you are spending quite a few iskies in the technology behind the cluster. It's encouraging to see that you are evaluating the layers of your architecture. The improvement in database performance through the use of RAMSAN has been a milestone in EVE, and it's allowed us to regularly see 30,000 players on. This is a huge increase from the early days of EVE when 3,000 players was a big number.

I'm troubled by your continued use of single physical server blades to host systems though. Even with dedicating a dual-CPU blade to Jita, you are limited to the capabilities, performance and stability of that single physical box. We experience it as a limitation on larger fleet fights, and the need for your admins to manually assign resources to hotspots. Also, the community seems to have some big issues with the process through which this the admins/GMs would get a heads-up of fleet battles from players.

I would suggest looking more seriously into virtualization. It seems that you're already leaning in that direction by having systems generate population statistics. Since you can assign a general statistic "weight" to a system's likelihood to see a population spike to the hundreds, you can currently allocate that system to boot onto a higher capacity blade. However, this isn't a dynamic model. You're constantly reacting to past behavior, and not current conditions. In addition, you're always limited by the capacity of a single CPU even if that CPU is at the momentary top of the heap. We also know that this bleeding-edge hardware doesn't come cheap.

If you were to virtualize with a set of tools like the ones from VMware, you could walk away from most of these problems. Your RAMSANS already do load balancing between each other thanks to the capabilities of SQL. Your SOL blades should be able to do the same. In essence, a SOL blade does the number crunching and chats with the database. You'll always have blades that are woefully underused (Jove space) and those that are within 80% of their theoretical maximum (Jita). This is obviously not an efficient use of your hardware. With virtualization, your entire cluster of SOL blades becomes one giant resource pool and the actual solar systems are instances running across this pool. With these types of capabilities you can create specialized high-performance resource pools for trade hubs, major fleet battles and events. These pools could run across higher-end hardware in the cluster so you could ensure optimum performance. The rest of the solar systems would continue to run on your older blades, but thanks to this resource pooling things would tick along as well or even better than they are now.

In short, I would recommend focusing more on virtualizing the SOL layer of the EVE cluster. This would make Tranquility flexible under dynamic load, it would lower the overall cost of procuring hardware, and it would improve performance under peak instances. You could use this capability to host more events, and even larger player groups.

What are your thoughts and comments? How do my comments compare with your vision for the evolution of the cluster?

Rawr Cristina
Caldari
Naqam
Posted - 2008.10.08 00:27:00 - [82]
 

Originally by: Mindstar

my node was equipped with the following...



who says CCP don't read the forums? Razz

this is why I loev them <3

oh! and nice blog. People will always complain about lag though. In 5 years or so you'll see people whining that they can't even have a 1,000 vs 1,000 without a bit of module lag... Neutral

Marlona Sky
Caldari Provisions
Posted - 2008.10.08 04:42:00 - [83]
 

So what is the likelyhood that an alliance would abuse such a feature as petitioning in where you 'plan' to have a big fleet fight only to go somewhere else??

Is there such an alliance that would do something like that??.... hmmm....

Hugh Ruka
Exploratio et Industria Morispatia
Posted - 2008.10.08 08:38:00 - [84]
 

Originally by: The Beekeeper
Nice blog, gotta love the techie stuff

Quote:
At peak hours, our database is processing over 2,000 transactions per second, which generates around 38,000 IOPS (input output operations per second)


This seems really quite low, those ramsans must be sitting around bored half the time, and I would have thought an enviroment such as eve's would generate way more transactions.


this ...

I am working around larger unix based systems and the disk arrays there have like 16 000 IOPs per port (the "low performing ones") with usualy 16 ports base configuration. I wonder why you need the ramsans. I guess proper IO load balancing and distribution on Windows is an impossible task :-)

The single threaded node process is a disapointment. You can only scale by CPU speed which is a waste. You can't even cram more node processes to one blade because you hit an IO wall. Per-grid threads maybe? So you can have at least SOME granularity ?

Doxs Roxs
Black Omega Security
Pandemic Legion
Posted - 2008.10.08 09:17:00 - [85]
 

Edited by: Doxs Roxs on 08/10/2008 09:17:34
That must be a spelling mistake, the Ramsans can handle 600 000 I/Os Per Second, each...

Regards
/Doxs


CCP Lingorm


C C P
Posted - 2008.10.08 09:25:00 - [86]
 

Originally by: CCP Mindstar
Originally by: DrAtomic
Aaaaanyway, did your wife move to Iceland as well?


While we did "meet" in a bar, I am not entirely sure I remember getting married.... I'll have to double check on that one I think.. Where do you get your information?? Very Happy

I think he got the In Crowds mixed.

Yes, my wife, did move to Iceland with me.

Cryten Jones
Gallente
Advantage Inc
The Matari Consortium
Posted - 2008.10.08 12:52:00 - [87]
 

Ok, i'll ask... what the hell is HPC world ?!?

-CJ

Crellion
Parental Control
Merciless.
Posted - 2008.10.08 13:18:00 - [88]
 

About a year into my eve experience I entered 0.0 alliance politics and fleet fights. It was a hoot!

Approximately 2 years after that I was driven out of 0.0 space by lag and my dislike for the experience. I still pew pew in 0.0 of course but not in fleet battles and anything that can be described as blobtastic.

Reading this blog I cosnidered goign back to all of that for the first time in the last year or two... and this is perhaps an apt measurement of the importance of the blog...

Keep up the good work.


Del ReyII
Caldari
Shadow Research Inc
Posted - 2008.10.08 13:30:00 - [89]
 

Originally by: Mashie Saldana
Originally by: Aprudena Gist
Originally by: Mashie Saldana
I'm curious how your Wolfdale blades will be configured as I have yet to see 4GB non ECC memory modules. The P45 and similar chipsets do support 16GB RAM however they only have 4 memory slots. So, do IBM offer socket 775 motherboards with 8 memory slots or have they developed 4GB sticks? I guess the other option is to run on only 8GB.


Well they are moving to 64 bit version of eve server to help with scalability so yes they would most likely do 8 or 16gb setups. They Intel Setup they have for jita currently is a 16gb one.

It's a hardware limitation with the Wolfdale as they use socket 775 and hence limited to unbuffered (non ECC) DDR2/DDR3 memory. The Woodcrest systems (running Jita) are Xeon based and uses ECC. The thing is ECC based systems are designed to handle a lot more memory, after all some Xeon motherboards can handle up to 192GB.


I'd say its more like LGA771 Wolfdale running on Intel 5000p using 4x 4GB FBDIMMs. Probably like a IBM BladeCenter HS21 .

Craleo
Caldari
Caldari Provisions
Posted - 2008.10.08 14:56:00 - [90]
 

Did you guys ever test VMWare for the nodes? You can take a four core node, install VMWare, install 1 instance of windows, and create one virtual CPU for that server. Then assign all four cores to that one virtual server, which only sees 1 CPU.

Would that work?


Pages: 1 2 [3] 4

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