Watch: Will Second Life's New AWS Cloud Deployment Improve Performance?

Linden Lab is starting to test a new deployment of Second Life running on the Amazon Web Services cloud, beginning with a copy of Blake Sea on the Second Life Preview Grid (known as Aditi)  -- and virtual world explorer Nodoka Hanamura just arrived from the sea with this video, putting the cloud through its paces in many vehicles. (Action starts at around 1:30 in.)

The most notable result? Near-seamless sim crossing, a much-desired but near-elusive goal. As you'll see in the video, there's a split-second pause when traveling from one sim to another, resembling what it looks like in many open world games -- as opposed to the several seconds or more rubber band-ing mess it's usually like in SL.

Frame rate, Nodoka tells me, what's fairly decent: 

"It varies from test to test but it's solid around the 20-30 FPS range in all tests on my specifications. With an average crossing time across all tests in the quarter to half second range." (Maybe not high enough frame rate for vehicle-to-vehicle combat, but certainly more than enough for a leisurely virtual yacht race.)

It is strange to me that there's any need for sim crossing (i.e. going from one server to another) when it's all in the cloud.

"It's mainly because of the fact that even if we had all the simulators on the same exact server or even same exact server farm at Amazon, data still takes time to go between client and server," Nodoka surmises. "It won't always be instantaneous, it's the same problem with cloud gaming.

 

"The only way you could hypothetically have zero delay is through," she adds wryly, "and this is a guess since I have no degree in it -- quantum-based time travel or some other way for packets to go faster than the speed of light." (So get right on that, Linden Lab.)

 

This is just one test, and results will probably vary greatly -- especially with multiple users, and especially if they are using the poorly optimized, high poly mesh avatars that are now so commonplace on Second Life:

 

"I don't see high poly avis really being impacted by simulator changes," as Nodoka puts it. "I mean, if the simulator and asset server are in the cloud I could see improvements in how fast that data is streamed to the user, but anything relating to graphics falls to client-side, and I think we both know Linden Lab won't touch the graphics pipeline with a bargepole lest all of Agni go up in flames."

 

Nodoka hopes to do Aditi sim-crossing tests with multiple vehicles, so if you're interested in joining her on that, follow her on Twitter -- and subscribe to her virtual world explorer group on YouTube.

Comments

[Alicia]

That's more than .2, .3 or even a half-second. Every time the person crossed over into the next sim, you could literally count past one. In fact, that's not even a seamless sim-crossing transition. It's another band-aid that will have problems once it starts to go live. Remember the mesh problem? It's going to play and factor into this, and not in a good way the second it goes live. Why? All the insanely high poly mesh avis will continue to factor into the loading and delay. Also, the engine and latency (as I noticed the latency of the person's video always in the yellow and red) will both play right into this which tells me that this has already failed.

Seamless sim/region-crossing requires an instantaneous transition from one zone to the next that's mainly seen in open world MMOs. The transition happens not even in a .2, .3 or even a half-second delay. It happens immediately and the transition is supposed to be smooth and easy.

[Nodoka Hanamura]

Alicia,

I never claimed it to be seamless. It never will be seamless. Also, did we watch the same video? Because I counted and even by more conservative estimates it came down to less than a second - and I checked using my video editor while working almost 6-8 hours on the blasted thing, and compared to going from sim to sim on SL's main grid - as I should know, being a veteran of SL Aviation and mainland exploration - This is a lot better.

I will not deny it's a bandaid, nearly every update to the code these days is, but expecting a 'instantaneous transition' from SL with how it was designed is like expecting Westboro Baptists to start tolerating.. anyone. It's not going to happen, and it won't happen unless the Lindens want to start over from scratch, which anyone who understands the intricacies of how a business works, knows will take years and millions of dollars in R&D, and not some simple 'just have the dev team fix it before focusing on anything else', or 'just put SL in a new engine'.

I'm optimistic, as is many of the people who have seen the footage, but I still say, even with my optimism that these tests do not readily compare to real life scenarios where the target or current simulators (or both) have a pre-existing workload. The only real life load that these tests can come close to is going from one sim to another in an area like the Blake Sea at non-peak hours or in open-water simulators such as those on the eastern border with Bellisseria and the Western border of Nautilus that connect the Bingo Strait to Corsica and Gaeta, with little to no domestic script load.

I would love to run tests with more realistic scenarios, such as having 30~ people in a target or destination simulator, or 5+ child avatars on a parent vehicle. But I don't have those resources readily at my disposal.

If we want the AWS transition to be as smooth as possible, we need to contribute to testing. Sitting here and being pessimistic and demanding the impossible (for SL) helps nothing.

[Pulsar]

Good job with the testing! This should be taken as an example about how to test and to report the results.
When someone just tells "it took X seconds/minutes to me", without specifying (at least) the network speed, the hardware and possibly the test conditions, it's of little of no help to replicate and compare the tests.
Imagine people debating about the time it takes to go between two cities: "it takes an hour", "nah, I took 15 minutes!", "meh, it was 45"... and nobody tells you if they used a the last Ferrari or a Ford Model T, on a country road or an highway and at which speed, if there was traffic, etc.

But this one is nicely documented.
Also, from the Viewer interface, I can see the drawing distance was set to 200 at 1:42 and correctly kept coherently the same among the various tests. It seems to me that too affects the region crossings a little, probably depending on what's around you, so using the same drawing distance is always good.
If one would be super precise, you could also state how many prims (not land impact) and scripts each vehicle has, and the scripts and the amount of attachments on you avatar, but I suppose the avatar was unchanged among the tests, although it could be interesting if other people do these tests on Aditi.

Of course there are all the listed caveats (no people on Aditi, the distance from the AWS servers etc) and also not everyone has a fiberoptic connection at 500+ Mbps - I know several people on SL with 12 Mbps A/DLS and even less - but it looks like quite an improvement, so I'm optimistic, too.

[Nodoka Hanamura]

Glad to hear the kind words, Pulsar.

On the topic of inconsitencies, the only difference in avatar was a change in outfit (merely of personal preference), but it was only one item, that being one of Ghoul's wetsuits.

Also, I'll be sure to keep in mind reporting true land impact (or prims, as opposed to indicated land impact).

I do hope that these results carry over to Agni, I honest to god, really do, because driving on mainland and flying are things that I love to do, and I really honestly want to be able to enjoy them with as little issue from crossings as realistically possible.

[Olivia Nova]

This is great news. I have sailing on my region, this will make it so much better.

[Tazgma Novs]

So second life is owned by Amazon lol. I glad that second life get hardware updates. Still be good

[Luther Weymann]

I'm thrilled at the results of this test. I spend much of SL time in a Dani floatplane or a BBX Design yacht flying or boating in the same areas as the test. To get any results near this test would be wonderful compared to the present. Although the other day I was flying in the North Sea and the region crossing was popping in less than one second. Very good video work, thank you for your effort at testing.

[Alicia]

@Nodoka, I actually did count and every time I did, I was able to get past one second every time. Now, add overly high poly mesh avis and objects, scripts, etc. and the performance will degrade.

[Wayda Dremascape]

Is getting better but not so good as what i try back in 2017 in Opensim Shin Grid

[Gery]

Linden Labs would do a lot more useful work fixing their servers and lowering the ping and improving their connection to mainland Europe. Having a 6 ms ping and 600+ Mbps connection, it goes to 190 ms ping and 110 Mbps when connecting to their servers, not to mention the huge packet loss (typically anything from 2 to even 100%!). It often takes me minutes for a low traffic sim to show. After almost 14 years on SL I'm really disappointed with how painfully slow it has gotten (in Europe at least) over the past 4-5 years and I'm ready to say goodbye to this mess that was once such a great way of connecting people around the world. Glad to see Linden Labs set their priorities straight... not...