Blake Sea is now up on Aditi!

[Fortnight Baxton]

I flew from Crow's Nest to Japan then back, before heading south into the empty 9 water sims before finally parking at Siren's Isle. Flew to my skill up to 70% speed in a S&W modded Bell 212 with 4 blades, floats active. Avatar Complexity of 38,004,  [14/14] running scripts, 800 KB allowed memory size limit, 0.059113 ms of CPU time consumed. Ran smooth as silk.

What's even more impressive is I've been maintaining a framerate that matches my current monitor. (75fps at worst), only losing 10fps when shadows enabled.

Edited August 1, 2020 by Fortnight Baxton

[animats]

Where's my boat?

Tested driving a boat into a region corner on an Aditi Blake Sea sim, and had a classic unsit and a lost boat. This doesn't surprise me; the double region crossing bug (BUG-214653) hasn't been fixed. If you start a second region crossing before the first was complete, the system breaks. Because the region crossings are faster on AWS, you have to hit the corner more closely and at higher speed to fail. But failure is still possible.

The odds of hitting this problem by accident on water or in air are low, but it comes up all the time on land, because there are roads that go through region corners. Which is why better land vehicles are region-corner aware and slow down. And why, in Bellessaria, there are buoys at water region corners.

A good fix is after a region cross, wait for the avatars to catch up. Normally that takes about 40ms, so you don't notice. That can and is done in some vehicle scripts, but it would be better to do it sim side. There was some concern from the Lindens that it would slow down region crossings, but in practice it doesn't.

Tested with forced network delay. 300ms and 500ms work OK, even for near region corners. 1000ms makes crossings near region corners fail, just like it did two years ago.

So, what I'm seeing is that faster networking and faster compute on the sim side makes the double region crossing bug less likely to come up. But it's not totally fixed. We'll have to see how it goes under load.

Single region crossings have consistently been good, even with forced network delay.

 

[animats]

The cloud is off today.

All the AWS sims on the beta grid seem to be offline right now. Non-AWS beta grid sims are still up.

[Lucia Nightfire]

2 hours ago, animats said:

The cloud is off today.

All the AWS sims on the beta grid seem to be offline right now. Non-AWS beta grid sims are still up.

Must have got the first bill from Amazon. 😆

[JoyofRLC Acker]

So here's an odd thing.  While this "yay cloud go cloud" euphoria is building here, back in boring old Agni, Blake Sea has been performing since Tuesday afternoon ... wait for it ... amazingly well.  Yes, that's right  - AMAZINGLY WELL.

Subjective observations of region border crossing freeze times in a Trudeau 12m yacht (NOT the slickest border crosser by a mile) are about a second (versus 5 - 7 seconds previously).  Chance of completing a race has gone way up to around 90%.   I havent fallen off in the last six races (3 on Shields, always reliable, and 3 on 12m).

Im not at all sure what's changed - supposedly Tuesday there was no server update, just "backend maintenance" (whatever that is; and which by the way caused total chaos during Tuesday morning, the asset server seemed to be running on molasses, very long times to logout and even longer to log back).

Can someone clarify what code versions are running on Aditi vs Agni?

[Nalates Urriah]

There are now a couple of videos to show people what the crossings look like. 

In this first one they guy doesn't know how to fly Shergood copters. Otherwise it is pretty good.

 

This one I made. I didn't have much experience with the Bandit 380R before I started recording.

 

 

Edited August 6, 2020 by Nalates Urriah
grammer

[Macrocosm Draegonne]

On 8/6/2020 at 12:29 PM, Nalates Urriah said:

This one I made. I didn't have much experience with the Bandit 380R before I started recording.

Those sim crossings are nice, well, all except hitting that dead region 😂 though thats not really any fault to consider lol.  I dont think ive ever seen sim crossings so smooth, this will make races of all sorts a lot better potentially.

Is it your graphics card glitching when the boat blinks in and out? ive never seen that before.

i used to work on real yachts/boats many moons ago, mechanical stuff usually, so im quiet fond of boats. I have to say thats a sweet boat, great design, sounds, particle effects, mesh and all, only caveat is its interior is a bit too low to the water level so you can see the water inside, and particle effects sometimes. too bad we dont yet have some sort of "occlusion" mechanism allowed to hide the water inside the boat for low riders, or even deep hulls, or submarines with spaces.

[animats]

On 8/5/2020 at 9:40 PM, JoyofRLC Acker said:

So here's an odd thing.  While this "yay cloud go cloud" euphoria is building here, back in boring old Agni, Blake Sea has been performing since Tuesday afternoon ... wait for it ... amazingly well.  Yes, that's right  - AMAZINGLY WELL.

The "fix" that affected the main grid has to do with objects, not avatars, crossing. So it looks like the big winners there are people with large vehicles. I'm not seeing any improvement on the main grid for motorcycles and a small helicopter. Still getting random fails just driving around.

The faster crossings on the beta grid AWS areas work fine.

If you're a vehicle scripter, please take a look at BUG-229231 and see if this would affect you. Thanks.

[JoyofRLC Acker]

6 hours ago, animats said:

The "fix" that affected the main grid has to do with objects, not avatars, crossing. So it looks like the big winners there are people with large vehicles. I'm not seeing any improvement on the main grid for motorcycles and a small helicopter. Still getting random fails just driving around.

The faster crossings on the beta grid AWS areas work fine.

If you're a vehicle scripter, please take a look at BUG-229231 and see if this would affect you. Thanks.

The improved performance seems to be persisting.   I'm measuring the "freeze time" at the border (as opposed to the agent detach / attach time), and that is regularly under 0.4 seconds now, usually under 0.2, versus 4 - 7 seconds before (on a Trudeau 12m yacht - 92 LI, 992KB script memory plus 400KB for HUDS).

The other thing that seems to have changed is that when the agent fails to re-attach, I'm finding that I can TP back on to the boat with about 90% success rate (cam is a little messed up and you're stuck in mouselook mode, but the boat is responsive to gestures etc) - previously the success  rate of getting back on was < 20% for me.

The bottom line is that we now have a very much higher % of boats finishing a race, and many more without falling off.

[Mollymews]

this just a general FYI for vehicle makers who might find their scripts now throwing errors under this new fast data transfer protocol way

i posted a mod of the Linden motorcycle script which shows the various techniques for dealing with region crossings. Old school techniques from back in the day, remembered by some, forgotten or never known by others

LSL techniques which when applied prevent all the LSL-thrown errors associated with the race condition of vehicle scripts and the permissioned agent being on different regions, and the old school way of providing a teleport facility for when the agent fails to transfer to the region when the vehicle does

script mod is here:

 

[animats]

7 hours ago, Mollymews said:

i posted a mod of the Linden motorcycle script which shows the various techniques for dealing with region crossings. Old school techniques from back in the day, remembered by some, forgotten or never known by others

LSL techniques which when applied prevent all the LSL-thrown errors associated with the race condition of vehicle scripts and the permissioned agent being on different regions, and the old school way of providing a teleport facility for when the agent fails to transfer to the region when the vehicle does

I do more than that, but I'm into defensive programming. What my bikes do on a region crossing is this:

When the changed/CHANGED_REGION event occurs, record the current velocity from llGetVel(), and turn off physics. This stops the vehicle hard. The usual delay is under 40ms, unless the crossing goes bad. So the user does not see this hard stop.

When all the avatars arrive (Molly's script shows a way to do that; look for "present")

• Turn physics back on
• Set the velocity back to what it was with llSetVel(). So you come out of the crossing with the velocity vector you had going in.
• Check for permission to set animations, and re-ask if necessary.
• Start animations.
• Reset camera.

If all the avatars don't show up in a reasonable time (Molly uses 10 seconds, I use 30 seconds, and i see 20 second crossing delays on Mainland about once per hour of driving), shut everything down and display a message. Giving the user a TP link to the vehicle is useful. If you can get your avatar back to the vehicle, you can usually get in or on and drive away normally. I once tried doing that TP automatically with RLV, but nobody liked turning on RLV just to drive.

How much of this we'll need after "cloud uplift" isn't known yet. We don't have a busy AWS sim to test on yet.

[Mollymews]

On 8/22/2020 at 6:41 AM, animats said:

I do more than that, but I'm into defensive programming

 

How much of this we'll need after "cloud uplift" isn't known yet. We don't have a busy AWS sim to test on yet.

more defense is generally better than not enough defense for sure

i take your point about poll time, sometimes it can take ages for the server to complete the transfer

the reason I tend to have a shorter lower polling time is that I find that I can sometimes get frustrated when I have to wait too long before attempting remedial action myself.  In my own personal vehicles I have a 5 second poll time. As in pretty much most instances we are broken if it takes longer than 5 seconds anyway.  And with a teleport slurl then we can attempt to recover sooner with a shorter poll time

edit add for completeness: as wrote the revisited motorcycle script has 2 polling periods.  The clock loop 'present' poll in changed event - for up to 10 seconds as wrote. Then it drops down into the timer poll for upto another 60 seconds as wrote.  Between the 2 polls the teleport slurl is offered.  In either polling period when the agent does arrive on the same region as the vehicle script then the reconciliation is attempted

on your second. Yes we don't know what Simon Linden might do as part of this.  Simon Linden the latest to have another go at fixing/ameliorating the race condition. I am hoping that the fix will be fully transactional. When the vehicle and avatar(s) do not complete transfer within time X, then the transfer is rolled back

failing a fully transactional transfer then what I would like is that when the vehicle script(s) transfer to the destination region then  the changed event is fired first (before all others, including the timer event)

knowing that changed (CHANGED_REGION) will always fire first then it will go a long way to making scripting vehicles a whole lot more predictable

Edited August 23, 2020 by Mollymews
add completeness