Tom Mueller interview 02 May 2017

[Robotbeat]

Seriously, fusion was just a complete side mention and contained no information about SpaceX*. Start a new thread.


*I know some will try to argue with that, but it's obviously true, so don't.

[feynmanrules]

I listened to about half of it so far but a big thanks to everybody who transcribed this thread in various forms. 

Saying so now because a few weeks ago a major Tesla researcher had a talk made public that was removed from youtube in a couple of days.    For spx- Good catch on saving the data, not 100% new but had some interesting tidbits which helps in understanding why this company does 1 thing vs another.

One thing that was a bit odd was that he referenced tesla per-unit costs less amoritization from upfront r&d + capex for model s.  It was from years ago and even then not a huge secret but also not in their sec data.   Might be considered material in some parts fwiw but I guess it's out now.  I'm glad the spacex specifics are being archived, esp the $/ton of o2 vs relative ch4 cost that mueller admitted was a bit of a misread on their part.   

My take away is that competitive diffierentiation is in space as most industries...   Who works the hardest for the largest value across the biggest segments of customers?    Spx seems to have more and bigger target markets than most incumbents and have approached them as if they were molding lego.   Should continue to yield interesting results for the company and hopefully grow the industry fast enough that it attracts a lot more investment.

[feynmanrules]

One other thing that was interesting which mueller mentioned which fits into NSF spacex constellation thread...

He was talking about 70% of your webbrowser is accessed via cache... which made me wonder... is one of the reasons the spx satellites is so big (400-100kg from memory) is that they're including massive cache-end storage systems?   (similiar to what google/akamai/etc already co-locate in a large #s of isps)

I have zero knowledge about the power/mass requirements of satellites for a given coverage area but I thought this was insightful as to their intentions.   Searched the threads for speculation from more informed folks but nada as yet.  Apologize if I missed it.

[Robotbeat]

I believe he said marginal cost for a Model S. That means, given their existing physical plant, etc, what is the cost for Tesla to make a Model S vs not making one. That should be much lower than the sale price, as it only includes marginal labor, materials, and maybe wear and tear on production equipment.

[Space Ghost 1962]

One other thing that was interesting which mueller mentioned which fits into NSF spacex constellation thread...

He was talking about 70% of your webbrowser is accessed via cache... which made me wonder... is one of the reasons the spx satellites is so big (400-100kg from memory) is that they're including massive cache-end storage systems?   (similiar to what google/akamai/etc already co-locate in a large #s of isps)

I have zero knowledge about the power/mass requirements of satellites for a given coverage area but I thought this was insightful as to their intentions.   Searched the threads for speculation from more informed folks but nada as yet.  Apologize if I missed it.

Two items - one is about CDN (content distribution network) caching which is location specific, and global caching (the sats are recirculating in multiple LEO planes, so if they "cache" and don't load/reload is they orbit continually, its for all on the globe). CDN content would either come out of the onboard global cache, or from say a metropolitan located CDN that temporarily relays through a passing sat (or sat chain) and down links.

In the first case, Google could price premium, high QOS content for CDN access, possibly to the highest bidder.

In the second case, Google could similarly price what is called peerage bandwidth of existing CDNs through Google's metro backbone to subscribers, also carefully priced.

It's a careful business strategy for global networks.

As to power/volume/other, look to SSD technology to get the basics on the amount one could fit in. Likely a few terabytes per each.

One benefit of being a LEO constellation is that like the ISS, one is under the inner Van Allen belts unlike geosats, so radiation isn't as much a consideration.

[IainMcClatchie]

"So why the heck does it cost some fraction of a million dollars to build a Merlin engine?"

Uh... it costs less than a million dollars?  That's pretty amazing.  That means the recovered stage 1 must be worth less than something like $10m.

[QuantumG]

Yeah, the idea that you could build a gas generator cycle rocket for less than $10M was once unbelievable. Any day now someone will produce one in their garage, although electric pumps have essentially taken that scale in a different direction.

[john smith 19]

As to power/volume/other, look to SSD technology to get the basics on the amount one could fit in. Likely a few terabytes per each.

Take a laptop SSD and de-rate using guesses on what level of redundancy SX would use.

One benefit of being a LEO constellation is that like the ISS, one is under the inner Van Allen belts unlike geosats, so radiation isn't as much a consideration.

Actually LEO still has some issues with the South Atlantic Anomaly

[Hobbes-22]

The comments from Mueller gives the impression that the M1D has hit all three items instead of just 2 in the cost, performance, reliability tradeoff. This is in itself very significant and also important for SpaceX. They have a very low cost engine with high performance and high reliability. Something that no one else in the US industry has even come close to. Which is why the Atlas V uses the low cost, high performance, high reliability RD-180 from Russia, the only other engine in use in the US that hits all three points.

This goal of hitting all three for the Raptor is also encouraging. Using lessons learned and piling the difficulty into development to be able to make such an engine instead of compromising on operational costs to keep development costs down. Musk has it right. If your going to fly lots of them and want to lower operational costs have good or high performance and high reliability, don't push off doing the right things in development for reducing the development costs.

I think hitting all three is the baseline standard...

What? The conventional approach is "cost, performance, reliability: pick two". Hitting all 3 is rare indeed.

Also, while the marginal cost per engine may be low, you have to amortize the development cost, and I get the impression the quoted low cost doesn't take that into account. This quote suggests development cost will have been high:

Musk convinced Mueller of using this method despite Mueller explaining what it is and how it increases complexity of R&D and increased costs due to blowing lots of hardware up before mastering the method.

What is conventional about EM's approach to problem solving?
Falcon design effort didn't settle for two, nor did Tesla... path chosen reduces to a value judgement of the person driving a development program.  Those that are driven by a committee usually settle for much less.

Hitting all three is only rare because most designers/manufacturers/committees don't try.

Note: The cliche is "Faster, Better, Cheaper -- pick two"

People don't try because hitting all 3 is generally seen as impossible. And it remains to be seen if SpaceX has done it. If they had to invest $1B to drive the marginal cost down to $30k, it'll take years for that savings to pay off.
SpaceX and Tesla aren't infallible, Tesla is finding out they're missing the "reliability" part of the triangle with the Model X, and Falcon 9 reliability is not above average for the industry.

[AncientU]

One other thing that was interesting which mueller mentioned which fits into NSF spacex constellation thread...

He was talking about 70% of your webbrowser is accessed via cache... which made me wonder... is one of the reasons the spx satellites is so big (400-100kg from memory) is that they're including massive cache-end storage systems?   (similiar to what google/akamai/etc already co-locate in a large #s of isps)

I have zero knowledge about the power/mass requirements of satellites for a given coverage area but I thought this was insightful as to their intentions.   Searched the threads for speculation from more informed folks but nada as yet.  Apologize if I missed it.

Nothing has been said yet about data cache as a function of the constellation.  Could be a potential new line of business for serving the ConnX and similar networks, since there is essentially unlimited, low cost power available* and a great heat sink.   Geosynch (not geo-stationary) orbits would be proper locations for these caches so that they can be content-specific and latency to GSO is irrelevant on content download.  When an appreciable portion of internet data are handled in space, there will be the full range of supporting services up there, too.

* Maybe this is one of those 'industries' that Blue Origins should be promoting to get off-Planet -- ITs in spaaace.  A significant fraction (10% and climbing) of the global electric generation goes to supply server farms.  Makes sense to generate electricity in space if you are also consuming it there.

the vast and electron-thirsty computer-server farms that make up the backbone of what we call “the cloud.” In his report, Mills estimates that the ICT system now uses 1,500 terawatt-hours of power per year. That’s about 10% of the world’s total electricity generation

http://science.time.com/2013/08/14/power-drain-the-digital-cloud-is-using-more-energy-than-you-think/

[Semmel]

Tom Mueller specifically talked about the backbone. That is the communication between the data storage centers, not the storage it self.

[AncientU]

Tom Mueller specifically talked about the backbone. That is the communication between the data storage centers, not the storage it self.

Right. 
All discussion and the FCC applications (latest attached below) only discuss backbone plus ground consumers.

Storage itself is someone else's problem (a.k.a., opportunity).

[envy887]

"So why the heck does it cost some fraction of a million dollars to build a Merlin engine?"

Uh... it costs less than a million dollars?  That's pretty amazing.  That means the recovered stage 1 must be worth less than something like $10m.

There is a lot more cost in a stage than just the marginal cost to produce the engines.

[oldAtlas_Eguy]

"So why the heck does it cost some fraction of a million dollars to build a Merlin engine?"

Uh... it costs less than a million dollars?  That's pretty amazing.  That means the recovered stage 1 must be worth less than something like $10m.

There is a lot more cost in a stage than just the marginal cost to produce the engines.

Currently it costs almost half the cost of the engines manufacture to test it singly in an acceptance test on the stand in Texas. A few $100K in costs per engine test.

But eventually if manufacturing of engines gets to the point of very few engines <.1% (1 engine out of 1000 manufactured) fails the acceptance test, the single engine acceptance testing could be eliminated and only the full stage acceptance test hot fire in Texas then would be performed and catch that 1 engine in 100 stages tested. So not only the test on engines individually would be eliminated but the shipment of the engines to and from Texas to Hawthorne would also be eliminated. Such that engines completing assembly move on in the factory to then be mounted to the stage. It could drop cost of engine manufacture and test by 20-30% over current. That would then drop cost of 1st stage manufacture which engines are 40% of the cost of the stage by  10% or more ~$2M.

[JamesH65]

The comments from Mueller gives the impression that the M1D has hit all three items instead of just 2 in the cost, performance, reliability tradeoff. This is in itself very significant and also important for SpaceX. They have a very low cost engine with high performance and high reliability. Something that no one else in the US industry has even come close to. Which is why the Atlas V uses the low cost, high performance, high reliability RD-180 from Russia, the only other engine in use in the US that hits all three points.

This goal of hitting all three for the Raptor is also encouraging. Using lessons learned and piling the difficulty into development to be able to make such an engine instead of compromising on operational costs to keep development costs down. Musk has it right. If your going to fly lots of them and want to lower operational costs have good or high performance and high reliability, don't push off doing the right things in development for reducing the development costs.

I think hitting all three is the baseline standard...

What? The conventional approach is "cost, performance, reliability: pick two". Hitting all 3 is rare indeed.

Also, while the marginal cost per engine may be low, you have to amortize the development cost, and I get the impression the quoted low cost doesn't take that into account. This quote suggests development cost will have been high:

Musk convinced Mueller of using this method despite Mueller explaining what it is and how it increases complexity of R&D and increased costs due to blowing lots of hardware up before mastering the method.

What is conventional about EM's approach to problem solving?
Falcon design effort didn't settle for two, nor did Tesla... path chosen reduces to a value judgement of the person driving a development program.  Those that are driven by a committee usually settle for much less.

Hitting all three is only rare because most designers/manufacturers/committees don't try.

Note: The cliche is "Faster, Better, Cheaper -- pick two"

People don't try because hitting all 3 is generally seen as impossible. And it remains to be seen if SpaceX has done it. If they had to invest $1B to drive the marginal cost down to $30k, it'll take years for that savings to pay off.
SpaceX and Tesla aren't infallible, Tesla is finding out they're missing the "reliability" part of the triangle with the Model X, and Falcon 9 reliability is not above average for the industry.

Impossible (to some definition) in the first round/deadline. Once you have two out of three, you can start working to get the third. Takes longer, meanwhile you have product up and working. Cost reduction is very common place in the mobile handset business for example, once the product has been out for a while. Seems to be the approach taken by SpaceX. Got it working, now working on getting it cheaper.

[RoboGoofers]

Impossible (to some definition) in the first round/deadline. Once you have two out of three, you can start working to get the third. Takes longer, meanwhile you have product up and working. Cost reduction is very common place in the mobile handset business for example, once the product has been out for a while. Seems to be the approach taken by SpaceX. Got it working, now working on getting it cheaper.

yeah but you're only optimizing the initial design. you can make it cheaper than version 1, but not cheaper than an engine designed from first principles to be cheap. 

[oldAtlas_Eguy]

I think you got it wrong. They started with cost and reliability for the M1D and have since been improving its performance. The costs have only marginally decreased from the beginning but the big advancement has been in thrust and performance (increases in ISP).

They needed low cost and reliability for a reusable engine. Having high performance was not a necessary item at the start with F91.1. But with 1.2 and now the upcoming Block 5 they have been concentrating on small cost savings small reliability increases and very significant performance increases.

[launchwatcher]

Nothing has been said yet about data cache as a function of the constellation.  Could be a potential new line of business for serving the ConnX and similar networks, since there is ... a great heat sink. 

Huh?   Vacuum is an amazing insulator.   

Multi-megawatt data centers on the ground can use the oceans and the atmosphere as a heat sink, using evaporative cooling, dumping heat directly into cold sea water, or other techniques.   Compared with the heat exchangers you can use on the ground, you'll need much larger radiators in orbit to sink megawatts into vacuum.

[cppetrie]

Huh?   Vacuum is an amazing insulator.   

Bingo. Hence the fancy engine bell on the MVac. Also, these!

[AncientU]

Nothing has been said yet about data cache as a function of the constellation.  Could be a potential new line of business for serving the ConnX and similar networks, since there is ... a great heat sink. 

Huh?   Vacuum is an amazing insulator.   

Multi-megawatt data centers on the ground can use the oceans and the atmosphere as a heat sink, using evaporative cooling, dumping heat directly into cold sea water, or other techniques.   Compared with the heat exchangers you can use on the ground, you'll need much larger radiators in orbit to sink megawatts into vacuum.

I understand the insulating quality of vacuum... work with it every day.  (astronomical instruments)

Just as it requires that one build and maintain giant solar panels, must also build and maintain large radiators.  Once built and orbited (assuming very cheap launch -- a few million dollars for a hundred tonnes per the interview) power then becomes free.  Server farms have HUGE electric bills, to the point that they are building their own power plants and even relocating to Iceland because electricity is cheap(er) there.

Iceland Lures Data Center Companies With Cheap, Renewable Energy

http://www.ibtimes.com/iceland-lures-data-center-companies-cheap-renewable-energy-2081695

Why Iceland Is a Hot Spot for Virtual Server Farms

https://www.engadget.com/2016/12/13/why-iceland-is-a-hot-spot-for-virtual-server-farms/