|
Author |
Thread Statistics | Show CCP posts - 25 post(s) |
|
CCP Yokai
|
Posted - 2010.08.25 15:55:00 -
[1]
Will post tech specs here shortly.
|
|
|
CCP Yokai
|
Posted - 2010.08.25 16:48:00 -
[2]
Seriously... only 1200... thought you guys had alts or something.
|
|
|
CCP Yokai
|
Posted - 2010.08.25 19:57:00 -
[3]
I thought if we posted about new hardware we'd get a few more people in local testing. Tell you what... we get this puppy above 1800 tonight I will post a vid of a dev dancing in a cabinet in the new London datacenter.
Or if you choose a server pic :)
|
|
|
CCP Yokai
|
Posted - 2010.08.25 20:03:00 -
[4]
Originally by: QioX Im wondering why is it only 1 quad core cpu and not 2 ? also why are you using a 10k rpm sas drive when these days most are using 15k rpm drives ? both of these would improve things alot more than just 1 cpu and an older slower drive
SSD's would not get even one more person in a fleet fight or Jita. The OS is there alone... no real exe's to speak of neededing the I/O there... so SSD's on a server with no IOPS... Just a waste.
Actually running a single CPU is Faster than running two CPU's when you are only using one core on one CPU. The CPU's talk via the QPI buss and create overhead and latency. For this test we want a clean benchmark... no other services are running save for Jita to prevent any overlap in serverload.
Hope that helps explain things a bit more.
|
|
|
CCP Yokai
|
Posted - 2010.08.25 20:10:00 -
[5]
Originally by: Teck Nine Edited by: Teck Nine on 25/08/2010 20:05:44 I thought they would have gone with AMD 6 core or 12 core server CPU rigs. I think the 6 core 4 CPU ports can suport up to or around 165GB of ram or at least use a 6 core CPU in the server rig.
I'm kinda disappointed in the new server specs tbh.
Again... this is the baseline server, not the production system. This is our benchmark... just need to see where she tops out to know how this would scale on more cores, RAM, and alternate clock rates.
Thanks!
|
|
|
CCP Yokai
|
Posted - 2010.08.25 20:16:00 -
[6]
Originally by: Textret
Originally by: CCP Yokai I thought if we posted about new hardware we'd get a few more people in local testing. Tell you what... we get this puppy above 1800 tonight I will post a vid of a dev dancing in a cabinet in the new London datacenter.
Or if you choose a server pic :)
We will require BOTH pictures.
Make it 2000 and I'll post both!
|
|
|
CCP Yokai
|
Posted - 2010.08.25 20:30:00 -
[7]
Exactly what he said :)
Also, the 130W CPU's are possible to use in the H series... just need upgraded power supplies. For now we only need to know what makes the most impact for each part of the server and this test will help.
;)
Originally by: Hawk TT
Originally by: Kephael
Okay... the spec:
IBM HS22 1 x Xeon 4C X5667 95W 3.06GHz/1333MHz/12MB 6 x 8GB PC3-10600 CL9 ECC DDR3 1333MHz VLP RDIMM 146 GB 2.5in Slim-HS 10K 6GB SAS HDD
Non-optimized and no affinity set. Intel tweaks will come after some counter collections get us stats.
As alwaysą tell us what you thinką if you felt itą if you want cookiesą etc.
Why no 5677, you guys getting cheap?
Because IBM HS22 and IBM BladeCenter H could not accomodate CPUs with TDP of 130W. Unforutantely Xeon 5667 3.06GHz is the fastest Quad Core with 95W TDP - the Xeon 5677 3.46GHz you've mentioned has a TDP of 130W...
Anyway, with Intel Turbo Boost feature (Xeon 55xx/56xx) the most loaded CPU core could go up to 3.73GHz (from 3.06), which is still more that the current 3.33GHz dual-core-4/5-generations-older Intel Xeons...Having better IPC (instructions-per-cycle), memory throughput and lower latancy, this node could bring 10-20% better performance (at least).
Let's see what the "Control Group" benchmarking will show ...
|
|
|
CCP Yokai
|
Posted - 2010.08.25 21:45:00 -
[8]
Just wanted to quickly say thanks!
We got over 1300 today... not bad for middle of the week. The hardware will stay there for now so there is plenty of time for testing.
Also, I wanted to point out that while I may have asked for more in people in Jita, it's all in good fun and in no way is anyone upset. I appreciate those who had the time and interest in logging on and respect those that couldn't or wouldn't. I hope that the next call for numbers will find you and yours available and interested.
We'll be doing some more testing shortly and will post in forums again when we are ready to try and break the record again.
CCP Yokai
|
|
|
CCP Yokai
|
Posted - 2010.08.30 13:02:00 -
[9]
Okay so here's the deal.
Testing is going REALLY well.
We tried the server on its default configuration to get a baseline. We have since disabled hyper threading... turned on turbo boost to max... and let the little hamsters take a few more pills.
Yesterday Jita peaked about 1550 players... with the server CPU hanging around at 80%.
Less than a month ago the CPU was at a near flat line 100% at just 1400 players. Part of this change is because of the super smart software guys making some changes (not going to steal their dev blog thunder) and then this new hardware.
Just has now been set to an 1800 limit from its previous 1400. I'd love to see that limit tested too. The goal is not to have a massive fleet fight in Jita. That's not a true indicator of the performance heuristics of Jita. But, if you can and if you are willing... please spend a bit of time in Jita today.
|
|
|
CCP Yokai
|
Posted - 2010.09.01 19:45:00 -
[10]
Originally by: Crieton What chassis do you guys use? BladeCenter H? Also, I know that blade supports 2 addins, have you used any of slots? I assume the FC backhaul card is installed but what about extra 1gbit ports?
Blade Center H
No need for fiber channel or extra gig-E cards on the server... none of that really helps. No disk I/O local or remote attached. And, the network throughput of EVE is very minimal.
|
|
|
|
CCP Yokai
|
Posted - 2010.09.01 19:47:00 -
[11]
Originally by: Zeba
Originally by: Jita Alt666 I heard that a ccp employee told a csm that we were allowed to attempt to break jita. That they simply wanted as many people in local as possible.
Is this true? Will I get into trouble for logging all my accounts on and sitting on all gates/undocks and insynch launching smart bombs?
For once Jitafest is an 'approved' activity.
We are asking playes to do what they normally do in Jita as much as possible not to skew the test results anymore than we need to. but some people do that stuff in Jita too :)
Jita is protected by a population cap (currently set to 1800), in the evet the server is over capacity we can simply reduce the cap to protect system stability. Breaking Jita completely is not... dare I say it... possible.
|
|
|
CCP Yokai
|
Posted - 2010.09.01 20:27:00 -
[12]
Originally by: Jita Alt666 You havent answered my question: Is it true a ccp employee has given the playerbase permission to attempt to break the Jita node? In other words before I contemplate this notion I want to have it in writing that I and others are not going to face bans for causing issues.
We've asked players to log on in Jita and play EVE.
If you do that and just that... there will be no issue. If you decide the best way to help us TEST or BREAK Jita is by means outside the EULA... then you do so at your own risk.
Does that answer your question?
|
|
|
CCP Yokai
|
Posted - 2010.09.01 20:55:00 -
[13]
Originally by: Jita Alt666
Originally by: Stick Cult Edited by: Stick Cult on 01/09/2010 20:39:49
Originally by: Jita Alt666 So the ccp employee who told a csm to: "Log in as many as you can in jita we challenge you to break it" needs to be censured by his employer.
No you're just twisting his words...
They want to test Jita by getting people in it. If it breaks, meh. If you're spamming lag inducing **** like there's no tomorrow, I bet they don't want that, as it wouldn't be an accurate test.
Quote:
CCP Yokai contacted CSM today with a challenge: Let's Break Jita Tonight!
As most everyone knows, Jita consistently places a heavy load on servers. Because of that, CCP began testing recent lag-related improvements to hardware and software in Jita. Test results indicate that Jita capacity has increased at least 25% compared to a month ago! The big question is: "Where's the tipping point now?"
yes totally twisted
There are no issues with the post. I asked the CCP employee to post that for me. It's quite clear we want as many players as possible in Jita. Do that :)
Thanks
|
|
|
CCP Yokai
|
Posted - 2010.09.03 09:50:00 -
[14]
Edited by: CCP Yokai on 03/09/2010 09:50:53 In response to questions about when do we do this for 0.0:
After Sunday we should have all the data we need to prove and measure the impact on Eve from:
Changes in Memory Latency Changes in Chip Architecture Changes in Cache Changes in Clock Rate Addition of Turbo Boost
When we have this we can much more easily spec the optimal system for Jita and similar systems and move on. This same server will be then used to benchmark fleet fights in 0.0
With all of your help the test in Jita gets us the predictable load we need for baseline data collection and we move right into improving 0.0
I hope that helps better explain the intentions of this testing.
|
|
|
CCP Yokai
|
Posted - 2010.09.08 11:08:00 -
[15]
Still collecting data and writing up the details. We are on it and plan to get something out in a Dev Blog format A.S.A.P
Give us just a touch longer please.
Thanks!
|
|
|
CCP Yokai
|
Posted - 2010.09.09 11:58:00 -
[16]
I can't promise sexy... but there are pics of... Me, CCP Explorer, servers, graphs and some words... Stay tuned.
|
|
|
CCP Yokai
|
Posted - 2010.09.10 16:24:00 -
[17]
New Dev Blog located here: http://www.eveonline.com/devblog.asp?a=blog&bid=792
|
|
|
CCP Yokai
|
Posted - 2010.09.11 11:49:00 -
[18]
Originally by: Jita Alt666 Dev Blog Summary:
Jita performance has reached a new benchmark level, testing of fleet fight performance will on same hardware will now commence.
I do like the sound of that.
Me too! I'm pretty excited to get a good fight on the new test hardware.
Thanks!
|
|
|
|
|