Hacker News Reader

Comment by ISL | Hacker News Reader

If you're in a car being held above the ground by a chain, every single link is a potential single-point of failure.

The classic counterexample to the chain (where each link is a SPOF) is the braided cable or rope, in which each individual strand shares the load collectively, and loss of some fraction of the total strands won't itself cause failure.

That's not to say that cables or ropes cannot fail, but when they do so, failure of many individual components is required.

A recent example is the failure of support cables for the Arecibo radio observatory in Puerto Rico in 2020. This was the result of a progressive failure played out over months, though at an accelerating rate, until the final catastrophic failure and loss of the instrument as a whole. (This wasn't wholly unforseen and was precipitated by a long period of rather intentional neglect of maintenance.)

Even after the failure of two complete cables, the remaining cables supported the load of Arecibo's instrument platform. On 1 December 2020, strands of one of the remaining cables began breaking, at an accelerating rate, leading to the total failure of that cable, then a second and third, and with it loss of the instrument platform. The final moments were captured on video.

Sequence begins at about 55s here: https://yewtu.be/watch?v=b3AASKr_iHc

Thank you for linking to one of the invidious instances. Youtube itself has become so annoying it's no longer workable for me.

What a fantastic comment. In a visual and immediate way to explain such a complex concept in one sentence.

A system may be implemted as a set of cooperating microservices each one being a (single) point if failure of the entire system.

Like many Node dependencies

Contrast a chain with double links, where if any link fails another is entirely capable of holding the load.

But the chain itself can be considered as one component and it becomes a single point of failure. Or if we go in the opposite direction, then we can breakdown links and say "Every single grain boundary is a single point of failure".

Partially-agreed on the former, but not the latter.

So long as the chain is not loaded near its tensile limit, then the grain boundaries that support the load within each link do so in parallel and are therefore redundant.

The selection of the allegory of the chain was intentional -- each link must be properly formed, or the entire chain will fail. If it breaks, it is surely correct to say, "the chain broke", but in truth, it was actually link-86.

For JWST, the remarkable/audacious thing is that many links in the chain from launch to observation are potential single-point failures. Furthermore, many of them haven't ever been tested independently in space... ever. It is a hell of a triumph that JWST has gotten this far already.

If even a small fraction of the instrumentation works at this point, we are going to learn a ton about the universe, simply due to JWST's position, collecting-area, and mirror-diameter.

You're right. I think I was searching for a better analogy. Abstractly: Sub-system SPOF conditions can be bundled up as a single System SPOF condition by multiplying the probabilities.

Also, I don't want to underplay JWST's success or its challenges. But, when saying 300+ SPOF conditions, one has to specify at what abstraction level. Otherwise, it can be misleading.

Each link in the chain is itself manufactured/created/constructed. If the weld is bad, the link may fail. "every grain boundary" is not formed in the same way each link is. The analogy is solid - actually, it's very good in that it conveys clearly and succinctly the concept the parent was asking about without introducing extraneous concepts. Very much in the flavour of EWD: "The purpose of abstraction is not to be vague, but to create a new semantic level in which one can be absolutely precise."

Another example is a small single engine airplane. The engine can be considered a single point of failure but if you zoom in it’s a complex system with many redundant components (multiple cylinders, multiple spark plugs per cylinder, etc.)

In a real chain, only one link is actually the weakest, and it is at that link where the chain will break, in the hypothetical you offer.

Which is true, but also not what's being discussed, which is "single point of failures", not "weakest link".

That’s not correct, someone with a bolt cutter can get any link in the chain and it causes collapse.

Single point of failure doesn’t mean “weakest link”. It means if this one piece (“single point”) fails, the whole system will fail.

But you don’t know which, and others may fail first due to various adverse events (say, random micrometeoroid strike). If any link fails, the whole chain fails.

Contrast a second chain, or double linked chain, so if any link fails the load is not lost.

This assumes the environment is completely predictable.

If you know the strength of every link in the chain with perfect accuracy, and you know that the only potential cause of failure is too much weight being placed on the chain, then the only link that can fail is the weakest[1] one because no other failure can happen before that one.

But really you need to design for the idea that various things might happen. Someone else gave the example of a person choosing a link to cut with bolt cutters. The person's choice is what's not predictable in that example.

---

[1] And if you assume it's not possible to have two links that are exactly as strong as each other.

"only one link is actually the weakest"

Which isn't relevant since (obviously) no one was talking about how many links are the weakest link.

"it is at that link where the chain will break"

Wars have been lost by thinking this way. e.g., D-Day was a successful assault on a strong link.

Reply on news.ycombinator.com