This Hacker News discussion centers around a new paper, “Optimizing Single-Threaded Applications with Exgen-Malloc,” and the broader trend of rethinking multi-processor system design. A commenter noted the work resembles existing “phkmalloc” without acknowledgement. The conversation then expands to the idea of abandoning Symmetric Multiprocessing (SMP) CPU coherency – essentially giving each core its own dedicated memory space and workload – to avoid performance bottlenecks. This is linked to projects like ByteDance’s “Parker,” a multi-kernel Linux approach, and other multi-kernel systems. While some see benefits in this approach, others express concerns about the increased complexity and potential for unpredictable behavior and security issues compared to hardware-managed cache coherency. The debate touches on historical attempts at similar architectures and the increasing relevance of locality as core counts rise, mirroring the performance characteristics of distributed systems. Ultimately, the discussion highlights a growing exploration of alternatives to traditional multi-core processing.