Massive data uploads since 2017.34 2448cfc

[chillaban]

Interesting... DOCSIS - Wikipedia I didn't know that.

I just ran a speedtest again.... No speed difference in downstream if uploading to my provisioned 50mbit.

The thing is, 50mbit down and 300 up is so easy for DOCSIS 3.0 (I have a 24x8 modem) that even if the modem sends one frame per direction at once, it's still your internet cap that is your bottleneck and not the capacity for the modem's inability to access the coax channels needed...

And FWIW, any speed above 50mbit where you start pushing DOCSIS 3 to the limit or needing DOCSIS 3.1 and beyond.... good luck getting a Tesla to do that, since it's a 2.4GHz 1x1 device.


At any rate, if you have 1mbit up DSL and a modem/router that does not know how to prioritize small packets in the upstream direction, yeah, uploads can make your downloads suffer. But if you're on a cable connection, that hasn't happened in sooooo many years given the way that cable modems enforce caps via traffic shaping queuing policies rather than a rigid cap.

 

[Xtek]

I've found my MX hasn't throttled. I observed it uploading at the full 10 Mbps I have available. Though it definitely seems to schedule the uploads. Since I had an ethernet drop put in my garage I plugged in a Ubiquiti AP Pro about 10 ft in front of the MX so it can get all the signal it wants.

By the way if you're having trouble with your S/X bouncing off of wifi right after connecting try putting it on your guest network or if you have the ability it's own network with VLAN. Having a DLNA device on the same network makes Tesla's wifi disconnect and I run a Plex server at home. Putting it on it's own wifi and VLAN fixed the problem.

 

[chillaban]

I've found my MX hasn't throttled. I observed it uploading at the full 10 Mbps I have available. Though it definitely seems to schedule the uploads. Since I had an ethernet drop put in my garage I plugged in a Ubiquiti AP Pro about 10 ft in front of the MX so it can get all the signal it wants.

By the way if you're having trouble with your S/X bouncing off of wifi right after connecting try putting it on your guest network or if you have the ability it's own network with VLAN. Having a DLNA device on the same network makes Tesla's wifi disconnect and I run a Plex server at home. Putting it on it's own wifi and VLAN fixed the problem.

Yeah, the DLNA interaction is pretty annoying and has been described a few times. That definitely helps to put it on a guest network. It's also worth playing around with your mirror auto-fold settings if the signal strength is marginal, because the antenna is on the passenger mirror and changes positions quite a bit whenever the mirrors fold. I have to turn it off in my parking garage in order to have a wifi signal.

 

[BigD0g]

You'd be surprised how many people are still running old old old modems. Friend of mine has 250/10 Mbps but has a DOCSIS 3.0 modem that is capping out at 50 Mbps.

But back on thread topic... up to 20.3 GB. I used AP2 quite a bit last night on 405 in Seattle. No major takeovers but it does freak out a bit when there are semi trucks in the next lane and it makes me nervous so I take over.

Does anyone know how much storage the X100D with AP2 would have?

At @verygreen can confirm, but I believe it's 32gb.

 

[verygreen]

Yes, it's 32G for ape + 8G for cid + 8G for maps + 8G for ic +4G for gateway.

That said the snapshots are limited to only at most 10G of storage.

 

[brkaus]

Learn such interesting things here!

 

[StefanSarzio]

I suspect Tesla is not throttling themselves but instead it's pure WIFI reception and 1 Mb is all they can manage.

No WiFi reception is fine.

 

[Cirrus MS100D]

Had our first (amazing!) family road trip from Bucks County, PA to Vermont and back in the Tesla. Lots of quality time on AP2 on Route 202, 287, and 87 in both highway cruising and stop and go (result of a big accident) traffic. IMO, it performed wonderfully and made the drive so much more pleasant, even enjoyable. (We've been doing this exact drive for about 18 years now, so we've experienced it in good traffic, bad traffic, good weather, bad weather, dangerously bad weather, day, night, etc.)

Got home and found our car was also excited to share pictures of the trip with her mothership: Uploaded 2.9GB between the time we got home (around 10PM) and when I checked the following morning. We are still on 2017.34.2448cfc.

 

[StefanSarzio]

I now saw upload speeds of about 7-8 Mb/s. Other sessions still stay below 1 Mb/s. No clear pattern recognisable.

Also when I come home the car mostly immediately starts uploading for some time, then stops, and later in the evening and again in the middle of the night starts uploading again.

 

[verygreen]

the late at night ones are definitely crashdumps. Car does not generate any regular snapshots while parked.

Do you have time of charge set ? Car wakes up for that.

 

[StefanSarzio]

No, I usually don't charge at home.

But due to Car no longer waking up from the app/API I have "always connected" set so that the car never sleeps.

 

[verygreen]

No, I usually don't charge at home.

But due to Car no longer waking up from the app/API I have "always connected" set so that the car never sleeps.

While the cid never sleeps, ape still sleeps a lot when parked. And it's the ape shutting down that currently generates those crashdumps.

 

[drewg123]

While the cid never sleeps, ape still sleeps a lot when parked. And it's the ape shutting down that currently generates those crashdumps.

I'm kind of amazed that they upload so many crashdumps that must be identical. It seems like what they should do is upload just the stack trace or md5 hash of the stacktrace, and then the mothership can decide if it is "interesting". That way, they can only upload N different copies of identical crashes. (sometimes multiple cores are helpful if you're tracking down a race condition, etc). Who knows, maybe they are trying to do that, and failing. Hmm..

@verygreen : Do they have enough debugging enabled to even get stack traces? Do their binaries have any symbols at all? Do they use ASLR? I suppose if they are doing ASLR w/o symbols, they may not be able to disambiguate traces easily...

 

[verygreen]

I'm kind of amazed that they upload so many crashdumps that must be identical. It seems like what they should do is upload just the stack trace or md5 hash of the stacktrace, and then the mothership can decide if it is "interesting". That way, they can only upload N different copies of identical crashes. (sometimes multiple cores are helpful if you're tracking down a race condition, etc). Who knows, maybe they are trying to do that, and failing. Hmm..

@verygreen : Do they have enough debugging enabled to even get stack traces? Do their binaries have any symbols at all? Do they use ASLR? I suppose if they are doing ASLR w/o symbols, they may not be able to disambiguate traces easily...

they don't have debugging symbols in shipped binaries, but no ASLR either. so just plug a core file into debugger against binary with debug symbols (that they do have somewhere internally I am sure) and off you go. Assuming they can get the full code file of course (my car somehow always aborts short, but it's possible it's working better for others since their uploads are much bigger than mine, eventually I'll run into someone with AP2 who is not averse to the idea and we'll see what their car uploads I guess).

 

[BigD0g]

they don't have debugging symbols in shipped binaries, but no ASLR either. so just plug a core file into debugger against binary with debug symbols (that they do have somewhere internally I am sure) and off you go. Assuming they can get the full code file of course (my car somehow always aborts short, but it's possible it's working better for others since their uploads are much bigger than mine, eventually I'll run into someone with AP2 who is not averse to the idea and we'll see what their car uploads I guess).

That would be me as soon as my stupid -ss cable arrives sir!

 

[verygreen]

That would be me as soon as my stupid -ss cable arrives sir!

I mean if you want me to root your ape for you, you'll need to visit me in person with the car in question. Or you can figure your own way of doing that too of course.

 

[pers1]

I'm now at over 16GB in 6 days in the Seattle area.

Over 18GB uploaded in the last week

PerS

 

[Puddles]

... I don't believe the mothership is doing anything for "anyones" specific route or conditions. They are simply uploading as many edge conditions and issues from the fleet and using that to enhance their algorithms. The neural net is not doing crap in terms of driving. the NN is simply object recognition at this time and lane markings and determinations ala comma.ai..

About this... I seem to remember a lot of chatter a while back about how feedback was refining driving maps, and the more people that were out driving, the better the mapping and AI response would become. Is it not like that anymore? Seems like the simplest way to improve the AP experience, at least from the point of simply staying in the lane.

 

[chillaban]

About this... I seem to remember a lot of chatter a while back about how feedback was refining driving maps, and the more people that were out driving, the better the mapping and AI response would become. Is it not like that anymore? Seems like the simplest way to improve the AP experience, at least from the point of simply staying in the lane.

At least from what @verygreen has said, so far the use of maps for driving in AP2 is basically nonexistent but starting to be deployed (e.g. the "atlas" thing).

I do agree that the use of maps would result in a huge gain in AP2 performance by not forcing AP2 to naively encounter every stretch of road over and over again and get it right the first time, and in that sense, a worse control algorithm paired with maps would probably outperform the current AP2….

 

[GregHudson]

Barely 10 miles of driving yesterday, I'm showing 600MB uploaded and slowly climbing with no driving done (yet) today.

About to embark on a 300mi roadtrip shortly, though. Looking forward to using AP2 on that boring stretch of 87 through New York to Albany, en route to Vermont.

What are you using to see how much data is being uploaded ?
Thanks, Greg