I kept having this sense of deja-vu as I read the article. This is essentially how the persistent store of the Apple Newton worked. Pretty neat.Our stuff supported vanilla battery-backed RAM, and we also supported memory-mapped flash (which has implications for how you represent the object state, since you can only flip bits in one direction without an expensive and destructive erase operation).Recovery after a crash was essentially a read pass of the whole device, which was fine for the storage capacities of the day, at memory-bus speeds.

Lock-free data structures tend to use the compare-and-swap atomic operation as a primitive: <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Compare-and-swap" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;Compare-and-swap</a>You provide three values to a CAS operation: the memory address you want to change, the value you expect to be there and the new value you want to be there. If the expected value is not what is at that memory address, the CAS fails; presumably, the expected value is not there because another thread&#x27;s CAS succeeded before you. You now need to do try again (and maybe do some more work before doing so).You can see this in Section 3.3, &quot;The insert operation&quot;: <a href="https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1909.02852" rel="nofollow">https:&#x2F;&#x2F;arxiv.org&#x2F;abs&#x2F;1909.02852</a>I think this answers your question: algorithms for data structures like this one use primitives like CAS to ensure that the data structure remains valid even when multiple threads are modifying it, and all without mutual exclusion mechanisms likes locks and mutexes.

The summary misses out on some key details that are most likely in the paper but are really needed to evaluate this.How do you ensure that two nodes are not initialised at the same place?

Efficient lock-free durable sets