|
Author |
Thread Statistics | Show CCP posts - 5 post(s) |
|
CCP Fallout
|
Posted - 2010.02.04 12:04:00 -
[1]
CCP Atlas' newest dev blog discusses our efforts to combat fleet lag, and shares some of the results of our investigations into the probable causes.
Please note: we are doing another test round on Singularity on Thursday, February 4, 2010 at 16:00 UTC. More information can be found in this thread.
Fallout Associate Community Manager CCP Hf, EVE Online Contact us |
|
|
CCP Atlas
|
Posted - 2010.02.08 09:37:00 -
[2]
Thank you all for your comments. I'll just make a brief comment about the concerns people have about the node being kept alive for the 1600 person fight that is mentioned in the blog.
We are deploying a fix to Tranquility this week (probably tomorrow) which identical to the methods used that fateful night in keeping the node up. The problem was that under great strain nodes were unable to contact the database to 'heartbeat' which is a way for them to tell the cluster that they are alive and kicking. A bug was introduced in the Dominion expansion where database calls were not scheduled 'fairly' so low level management functionality such as the heartbeat would not have enough precedence over game functionality such as shooting a gun. This was jury-rigged live on Jan 28th and is making its way to a permanent fix this week. The fix that is being deployed is identical to the live-fix that was made on Jan 28th.
So, despite 'making people faint' comments this change was done in an extremely controlled manner and was not a case of 'keeping the node alive longer than it should have been'. If the node would have been allowed to die the results would have been extremely bad for everyone as the solarsystem would not have been properly persisted, and it would have come up on a different node, where it would immediately die again because of the population count and that would continue to cascade as long as there were more than ~1000 people fighting there leading to horrible gameplay for those involved and a possible destabilization of certain parts of the cluster.
The cluster cannot gracefully handle a 1600 person fleet fight right now but bouncing between dying nodes every 2 minutes with the solar system in an indeterminate state is not really any kind of a solution.
We will be deploying fixes this week and the next and are continuing to closely monitor fights as they occur. Please bear with us a little while longer.
|
|
|
CCP Atlas
|
Posted - 2010.02.08 21:45:00 -
[3]
Originally by: Dratic Good to see things are being looked at to get back to pre dominion numbers in fights. What i'd like to know is if there are any updates on stuff like HPC infiband to improve preformance, and ultimately using more than one CPU core for a system. 1600 just isn't enough as the game continues to grow. The removal of aoe titan doomsday really didn't make your job easier as the hordes can now enter and not get wiped on jump in (barring the enormous camp, lag node crash stuff).
Low level optimizations and ways to better spread the load is actually something that is being actively worked on by our Core Cluster group and hopefully we will be able to push the boundaries further so that Eve can support the fleet fights of the future.
|
|
|
CCP Atlas
|
Posted - 2010.02.08 22:26:00 -
[4]
Originally by: BlackHorizon ...from what I can read from CCP Atlas' posts, CCP forced the node to stay online.
That is actually not what happened. A bug which caused nodes to die was fixed. It is as simple as that.
|
|
|
CCP Atlas
|
Posted - 2010.02.09 10:16:00 -
[5]
Originally by: Allen Ramses ... On January 28, CCP disabled safety protocols (emergency warp, despawn, inter-node synchronicity) to keep the node from dying. A few days later, CCP Atlas states that it was a fix for the bug that was causing the node to crash. ...
That's pretty far from what happened... A fix to a bug introduced in Dominion that caused nodes to die was deployed live. The only effect of that fix was that the node did not die but recovered eventually as it would have before the Dominion expansion.
If the node had been allowed to die due to this new bug the solar system would have ping-ponged around the cluster as long as people would have been logging on to it and fighting in these great numbers. That means that it would have come up on one node, the node would have died. It would then come up on another node and it would have died as well bringing down the other solar systems on those nodes which would get remapped. This, in turn, could have destabilized the whole cluster, so this was a pretty nasty bug and needed to be fixed then and there.
Eve has never been able to support anywhere close to 1600 person fleet fights and we have never claimed to support those numbers. Hopefully we are heading in that direction with some of the optimizations that are in the works but we're not there yet.
|
|
|
|
|