As discussed in the Advent of Code 2024 thread
viewtopic.php?p=2314032#p2314032
the three black kittens have been challenged to write a parallel solution to day 24 that runs on the super cheap cluster. As that's this thread, I'll post the results here.
To begin Scratchy copied the Go version of the day 24 solution and ran it on the head node. I find it irritating when kittens run computations on the head node, but as the dog developer is no longer around, I just ignored it.
The result wasThis shows that the Pi 4 is
548.01 / 4.6935 = 116.76
times faster than the original Pi. Since there are five computational nodes in the cluster, best case scaling would lead to a solution that still runs about 24 times slower than the Pi 4.
This highlights the fact that the goal behind the super cheap cluster is not to build a fast machine, but to create an environment in which to learn MPI distributed-memory parallel processing. To this end the next step is translating the Go program to C, C++ or Fortran. I'm hoping the kittens choose C as that is the language I am most comfortable with.
viewtopic.php?p=2314032#p2314032
the three black kittens have been challenged to write a parallel solution to day 24 that runs on the super cheap cluster. As that's this thread, I'll post the results here.
To begin Scratchy copied the Go version of the day 24 solution and ran it on the head node. I find it irritating when kittens run computations on the head node, but as the dog developer is no longer around, I just ignored it.
The result was
Code:
$ ./go24 # Pi B+ 700 MHzAdvent of Code 2024 Day 24 Crossed Wires (GOMAXPROCS=1)Part 1 The z wires output 568353998Part 2 Swap wires brp,bwv,dmk,dph,jbj,jnf,nfk,pgkTotal execution time 548.010919033 seconds.548.01 / 4.6935 = 116.76
times faster than the original Pi. Since there are five computational nodes in the cluster, best case scaling would lead to a solution that still runs about 24 times slower than the Pi 4.
This highlights the fact that the goal behind the super cheap cluster is not to build a fast machine, but to create an environment in which to learn MPI distributed-memory parallel processing. To this end the next step is translating the Go program to C, C++ or Fortran. I'm hoping the kittens choose C as that is the language I am most comfortable with.
Statistics: Posted by ejolson — Thu May 15, 2025 4:46 am