Time Dilation – How’s That Going?
Since Time Dilation (TiDi) was activated fully on Jan 18th, we’ve seen it trigger in many different places, from Jita to some of the biggest spaceship slugfests we’ve seen in some time. In all cases, it has kicked in appropriately when the server node has become overloaded, keeping things running responsively and sanely.
I want to share my favorite two graphs of the past few weeks with you folks. They display the amount of time dilation in red against the right axis with how delayed module processing is in blue along the left axis, in seconds.
F-9F6Q on 20Jan: >1,300 pilots on a reinforced node (4 solar systems, supernode hardware)
92D-OI on 21Jan: >1,350 pilots on a normal node (83 solar systems)
The thing to notice here in both cases is that modules were not allowed to lag very much. When the delay spikes up, TiDi kicks in, then the delay goes away. Happy news for anyone interested in making lazors go pewpewpew or reppers go, well, whatever they go. In both huge fights here, the module response time was kept under one second for the vast majority of the action, which is a tremendously large improvement over the 20, 40, 600 seconds we’d sometimes see in fights of this scale.
Another thing the attentive amongst you will notice is the large pile of load at the end of both fights, corresponding to large jumps out of the system. Jumping being expensive is one of the first things we identified back when Team Gridlock was formed, so we’re very familiar with that problem and why it exists, but it’s not easily dealt with. This load has always been there, it’s just now very explicit and visible to you folks where before you’d just notice it by sometimes being black screened or at e-warp distance from gate or other such tomfoolery. In any case, the work to be done to remedy this has been identified and is awaiting being the worst problem we have so we can tackle it.
While I’m here, I want to address another common criticism of TiDi – that it’s a node-level thing, which causes solar systems potentially far away from the fight to be effected. Ideally, we could have a clock per solar system and just slow down the biggies and that’d be just swell. Unfortunately we’re very far from there just based on how our software is set up, there would have to be some pretty fundamental changes in order to keep track of all that. Given the choice between not doing TiDi and doing it on a per-node basis though, I believe I made the right call.
But enough about all this negative stuff. TiDi is going quite well and big fights have been quite enjoyable as late. Or so I hear. I actually haven’t been in any personally, I just watch. Let me defer to Lovelocke's exhilarating battle report from EVE Swarm about the battle of 92d:
"As an estimate I believe we had around 650 – 700 people, not including capital ships. Understand that this is an approximate figure – it was far too difficult to pinpoint an exact amount due to a mixing of friendlies and hostiles in local. I have to be honest and say I expected to be ‘blue balled’ as such an amassing of sub capital ships would surely put most people off.
I was wrong, very very wrong. A cyno went up in system and local began to spike. “Spike” doesn’t sufficiently explain by how much local went up in so much as local actually more than doubled! Levelling off at around 1350 – 1400 the TiDi indicator quickly turned red..."
"...The scene was magnificent; to the left were well over 30 Raiden titans and many more super carriers that were saving their Tech moon tower, up above were Goon Titans doomsdaying the lone trapped Erebus that had been surrounded with bubbles and all around were hundreds of Maelstroms fighting a fierce battle against the Drake horde. As the Dreadnoughts were sieging Raiden Titans began doomsdaying them in a horrifying yet beautiful lightshow of brilliant greens and vibrant yellows, followed by massive explosions. A calculated sacrifice.
This, my fellow readers, was the opening minutes of a battle that would rage for approximately three hours..."
"...We carried on fighting, many of us died and we killed many. But do you know what was most remarkable about this fight? The lag was barely noticeable. 1400+ in local, super capitals with fighter bombers swarming, doomsdays, Drake missiles, etc, yet with TiDi in full force there were no crashes, no unresponsive modules, no MWD’s that wouldn’t turn off. Astonishing. I still don’t fully understand what TiDi is, its explanation far too technical for my fragile little mind, but whatever it is CCP have certainly struck gold. This was by far a huge improvement over even smaller battles I’ve encountered in the past."
Right then. Going to chalk this up firmly in the “win” column. Next up – we’re hitting the client! Expect some pretty bigtime improvements there in the coming months. You can check out the first round of them right now over on Sisi! There's also some information on the Crucible feature page.
This shows the cluster-wide average amount of time dilation for the given time on the clock for the period mentioned. Unsurprisingly, most of the usage sits in the EU & US primetimes. The spike down at the end there is our new TiDi-enabled shutdown routine, which pauses the universe while shutting down. It’s kinda neat like that.
In that same period of time, 12.87 hours of simulation time in total have been dilated out of existence – 45 minutes per day on average across the entire universe.