XPost: sci.math, comp.lang.prolog   
   From: janburse@fastmail.fm   
      
   Hi,   
      
   The Linux kernel's RCU-protected hash tables   
   is still not free of problems. Multiple writers   
   might still need extra work.   
      
   String interning with a HashSet (no values, just keys)   
   is actually easier to make lock-free without full   
   CAS because you're only concerned with existence,   
      
   not updates.   
      
   Bye   
      
   P.S.: I am not affected by this amplified nonsense.   
   Dogelog Player has even no atom table. And since it   
   is single threaded, the HashMap for predicate lookup   
      
   is totally lock free. It is single threaded and it   
   has cooperative multithreading. Internally it uses   
   async/await from JavaScript for example, which is   
      
   a cooperative multthreading approach, but from the   
   outside it provides tasks and sleep and stuff.   
   Recently demonstrated a little Strudel style music coding:   
      
   Strudel Coding in Dogelog Player   
   https://medium.com/2989/bbb9c78fcd67   
      
   In basically declared multi-threading dead, before   
   AI accelerators were there. I had more some worker   
   things in mind, with more thorough siloing and isolation.   
      
   But now that we have AI accelerators, the decision   
   to abandon multithreadinhg looks even more splendid.   
      
   Mild Shock schrieb:   
   > Hi,   
   >   
   > An example of a lock free datastructure, that   
   > even doesn't use CAS, is for example:   
   >   
   > Read-Copy-Update (RCU) Based Hash Tables   
   > These use only memory barriers/fences and atomic pointer writes:   
   > - Basic approach: Readers access the table without locks,   
   >    writers create new versions   
   > - Memory reclamation: Uses RCU grace periods instead of CAS   
   > - Example: Linux kernel's RCU-protected hash tables   
   > - Operations: Only requires atomic loads/stores and memory barriers   
   >   
   > For Prolog systems there are also various   
   > approaches arround, if one aims at the multi-threading   
   > model for dynamic databass or atom tables.   
   >   
   > I think this multi-threading model should be   
   > abadoned, in favor of things that can be speed   
   > up by a AI accelerator. Dogelog Player has abandoned   
   >   
   > multi-threading all together. But for example   
   > SWI-Prolog has heavily focused on lock free   
   > data structures already like 10 years ago,   
   >   
   > and it seems YAP can still not keep up with   
   > SWI-Prolog. See for example here:   
   >   
   > Yet Another Lock-Free Atom Table Design   
   > for Scalable Symbol Management in Prolog   
   > https://link.springer.com/article/10.1007/s10766-024-00766-z   
   >   
   > But in my opinion, in the light of the AI Boom,   
   > this is all amplified nonsense.   
   >   
   > Bye   
   >   
   > Mild Shock schrieb:   
   >> Hi,   
   >>   
   >> What are you, a 5 year old moron?   
   >>   
   >> There are millions of algorithm that use volatile   
   >> variables. Just look at the Java code base.   
   >>   
   >> But I was not refering to multi-threading, I   
   >> was refering to PRAM for matrix operations.   
   >>   
   >> See for example here:   
   >>   
   >> Hogwild!: A Lock-Free Approach to   
   >> Parallelizing Stochastic Gradient Descent   
   >> https://arxiv.org/pdf/1106.5730   
   >>   
   >> Fuck off moron.   
   >>   
   >> Bye   
   >>   
   >> Blending Molostvov schrieb:   
   >>> Mild Shock wrote:   
   >>>   
   >>>> What are you, a 5 year old moron?   
   >>>>   
   >>>> Pascual Sokolsky schrieb:   
   >>>>> Mild Shock wrote:   
   >>>>>   
   >>>>>> But in principle the architecture is rather:   
   >>>>>>   
   >>>>>> parallel random-access machine (parallel RAM or PRAM) is a   
   >>>>>> shared-memory abstract machine.   
   >>>>>> https://en.wikipedia.org/wiki/Parallel_RAM   
   >>>>>>   
   >>>>>> The above class of machines is not widely know.   
   >>>>>> But PRAM has been also studied, already in the 80's.   
   >>>>>   
   >>>>> parallel read of shared memory is only allowed to cia and the chinese   
   >>>>> governoment; must be somenthing you dont know   
   >>>   
   >>> from shared memory you only read and write sequential, me frendo,   
   >>> driven by semaphores, atomic instructions and so on. You are not that   
   >>> fucking stupid to write parallel to a cell, are you   
   >>>   
   >>   
   >   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)