> Is the cost of projects such as Webb so high because of high launch costs In my head (and I totally have nothing to do with NASA/JWT), it’s the reliability r

> Is the cost of projects such as Webb so high because of high launch costs In my head (and I totally have nothing to do with NASA/JWT), it’s the reliability requirements. I have this discussion with clients a lot, I had it just last week. They’ll come to us with a project with some vague requirements and I’ll ask “what are your reliability requirements here?” and very often they claim “this is business critical, it needs to be 100% reliable”. So I go into my spiel about how 100% is not an achievable goal, and we need to be realistic about whether their actual requirement if 99.9999% or 99% or somewhere in between. For web/app-backend/*aaS type projects I’ll point out I can provide 99% SLA hosting for around $100/month, but that each extra “nine” costs ten times more. If you want guarantee less than an hour of downtime per month, it’ll cost you $1000/month to host, if you want less that 5 minutes a month of downtime I’ll charge 10,000. If you need less than 30seconds a month of downtime we’re talking an on-call team of 5 engineers on 24x7 rosters and that’ll cost you $100,000 a month. If I were launching a one off custom built billion dollar telescope to an orbit a million miles away with practically zero chance of ever fixing anything that might need “hands on” or replacement, I’d be looking very carefully at the costs for 6, 7, or 8 nines of reliability. And I’d choose to spend a _lot_ more than most people would consider “reasonable”. There’s a difference between monthly web hosting costs and the JWT reliability, but instead of 10 or 100 million a month over the life of the project, they need to spend a similarly astronomical budget on the architecture, design, manufacturing, and testing all up front. Can you imagine designing a web hosting platform where the requirement was to run for 5+ years with zero downtime and zero human intervention? U can’t “more hardware” your way out of reality there, and you can’t ignore the risk of your cloud hosting going down (or under). So you now need to think along the lines of 8 nines of HA platform with exception handling and self healing for every possible scenario, in a multi cloud configuration in case Amazon goes into liquidation, that is resilient to _everything_ from as yet undiscovered OS and hardware vulnerabilities to DNS and SSL cert outages and BGP hijacks. (That’s turning into a fascinating thought experiment for me now. I wonder how I’d plan things if someone said “I’ll pay to 1 billion dollars if you build me a (non trivial) backend hosted platform/application, that has less than 1 second of downtime in 5 years, and the _only_ maintenance you can do is reboots or software updates over a 300 baud dialup modem, no platform/hardware configuration changes allowed..)