Hacker News Reader

Comment by mandeepj | Hacker News Reader

To help you with not feeling overwhelmed - they are doing it since May 6, 2002. It has taken a lot of time, of a lot of people to get here. There was a lot of failures also, along the way.

View on HN

A lot of what they are doing dates back a lot further than that. The innovations SpaceX has brought to the industry are more process related than technical. SpaceX has done a great job of setting a streamlined vertical integration for their production processes, which has allowed them to iterate a lot faster than others in the industry.

That isn't to say they haven't been doing interesting things from a technical perspective (they're landing rockets on barges, after all), but the real secret to their success has been taking proven technology and doing it better/more efficiently (for instance, the fundamental design of their rocket engine, the Merlin, dates back to 50's).

So, out of curiosity, was there one person responsible for SpaceX operating in this manner? Was it Musk, or someone else?

Unless they accomplished a miracle by building a world-class engineering team from zero, there must have been a catalyst.

Elon Musk definitely deserves a huge share of the credit for SpaceX's success, but there have definitely been other folks who have made (and continue to make) critical contributions.

Just a couple examples would include Tom Mueller, who was responsible for most of the design of the propulsion systems SpaceX currently uses (including the Merlin and Draco engines), and Gwynne Shotwell, SpaceX's President and COO. SpaceX absolutely owes its existence to Shotwell, both its commercial success, and the fact that it didn't die (a number of times) are directly due to her.

From an outsider's perspective (_very_ outside), it seems like it's really the "synergy" of Elon and Gwynne that have made SpaceX what it is today.

> there must have been a catalyst

There is - having visionary, charismatic founder(s). This pattern has been successful at Apple, Microsoft, Amazon, Facebook, Netflix, Tesla\SpaceX, Google.

The resulting cost savings they (SpaceX) are bringing to the launch market is really amazing. United Launch Alliance (ULA) made a fortune on over-charging the United States Government for their National Security launches. SpaceX is now competing for those launches with a price between 1/4 to 1/3 of what ULA is charging.

Part of that also comes from the pricing model. ULA has often done cost-plus contracts, which incentivize inflating the cost. When SpaceX joined the party, most of the contracts became fixed price (i.e. the government is going to pay $X, and it's up to the launch provider to build their profit into that, and any cost overruns are the provider's problem).

> SpaceX is now competing for those launches with a price between 1/4 to 1/3 of what ULA is charging.

Is there really a competition for this much price difference?

I remember reading here a document about C programming guidelines for military / space software ( basically no dynamic memory allocation ever), but that was set a long time ago. I wonder if spaceX reinvented some parts of the software stack, or if they went along the traditional way.

I don’t expect seing rust code in the shuttle, but maybe formal proof with custom functionnal languages...

From what I understand, they are pushing the boundaries on the hardware front, i.e. relatively recent, typical Linux computers onboard, and just relying on having a few redundant ones and short flight times, rather than very radiation hardened systems like those used in satellites or real-time operating systems.

However on the SpaceX software side I think they’re fairly conservative and “boring”, so I’d guess C and similar guidelines to what you’ve mentioned.

I'd be surprised if they were able to be human flight certified without all the redundancy and hardening that any other manned spacecraft has, though.

For example, my understanding is that the redundancy approach is basically "do everything on 3 computers, vote on the outcome".

Sometimes you even have different firmware/software teams write code for each of the three different configurations (often targeting different processors/architectures, etc). Pretty cool

You can "skimp" on hardening if you use more redundancy (which is generally cheaper, and often lighter (which is the real cost savings)).

You can also skimp on hardening if the operational lifetime in space is minutes to days, instead of months to years like most satellites.

They build everything in-house - software and hardware. You can read about it in detail in Musk's biography - https://www.amazon.com/Elon-Musk-SpaceX-Fantastic-Future-ebo...

The flight software team did a reddit AMA a while back [0]. They use C++ for writing the flight software (what flies falcon, dragon, etc) running on a customized RT Linux.

[0] https://www.reddit.com/r/IAmA/comments/1853ap/we_are_spacex_...

And FreeRTOS. RTLinux is only barely soft realtime.

> RTLinux is only barely soft realtime.

Linux used to run as a non-privileged process in hard real-time kernel, which taken together was guaranteeing responses in nanoseconds range.

If this is still the case (along with what Wikipedia states) , the whole system is by definition hard real-time.

But only the underlying kernel part is actually real-time, the Linux process wouldn't be, right? That leads me to the question: Shouldn't all things in a rocket be real time? Are they "post-poning" stuff like telemetry and other communications?

All critical processes can run outside of Linux, in hard real-time fashion. Linux is used for non-critical things, like networking... The whole RTLinux includes Linux (non-RT part) and a hard RT part, for which you can write your code.

Wow, that's really silly that someone went out, made a hard realtime system using Linux as the lowest priority level, and then called it RTLinux. RTLinux for years had meant Linux with the PREEMPT_RT patches.

In any case, SpaceX doesn't use Linux for hard realtime systems.

> They do not have hard realtime requirements

https://lwn.net/Articles/540368/

And I probably wouldn't use the RTLinux for a real hard realtime system if I were making the choice. Putting all of the realtime code in the same memory domain as the rest of the Linux kernel is a huge surface area for corruption. I'd probably use sel4 with a virtualized Linux to get the same semantics.

Thanks for this! ”We’re looking to start crew trials as early as 2015.” :)

Reply on news.ycombinator.com