I haven't really kept up with the literature, but back in the late 2000s I spent a lot of time reading academic papers from the concurrency and distributed systems community. Back then, many...
I haven't really kept up with the literature, but back in the late 2000s I spent a lot of time reading academic papers from the concurrency and distributed systems community.
Back then, many researchers were following the hype train of Software Transactional Memory (STM). Note that "software" in that context was kind of a misnomer: What they were actually trying to do was to achieve transactional memory while avoiding any new additions to instruction set architectures. Instead they were working toward building transactional memory out of existing hardware primitives such as atomic double-width compare-and-swap.
There is clearly a divide between academia and practitioners. For example, Cantrill laments, "no data, no qualifiers, no study, no rationale, no evidence of experience" to back up assertions like locks not being composible, but the academics would never even consider that because to them it is mathematically proven.
OTOH, academically-developed concepts such as linearizability are essential for proving the correctness of lock-free algorithms.
I haven't really kept up with the literature, but back in the late 2000s I spent a lot of time reading academic papers from the concurrency and distributed systems community.
Back then, many researchers were following the hype train of Software Transactional Memory (STM). Note that "software" in that context was kind of a misnomer: What they were actually trying to do was to achieve transactional memory while avoiding any new additions to instruction set architectures. Instead they were working toward building transactional memory out of existing hardware primitives such as atomic double-width compare-and-swap.
There is clearly a divide between academia and practitioners. For example, Cantrill laments, "no data, no qualifiers, no study, no rationale, no evidence of experience" to back up assertions like locks not being composible, but the academics would never even consider that because to them it is mathematically proven.
OTOH, academically-developed concepts such as linearizability are essential for proving the correctness of lock-free algorithms.
EDIT: Made a clarification about STM