gparmer / forking_pennycoins

Fork, Communication and Pennycoin Mining

Heard of bitcoin? Dogecoin? You haven't seen anything until you've seen the new PENNYCOINS!!! We wrote a program to use many processes to mine pennycoin for us! We'll be rich!

Problem is, we have a performance issue. We're running the mining in many different processes, and they all finish at different times, but they take forever to get back to us. We want to get the pennycoins back as they are mined, and don't want to wait for the next one to cache in. Currently, it does this:

church$ ./bin
Mining 0 done, cha-ching!
Mining 1 done, cha-ching!
Mining 2 done, cha-ching!
Mining 3 done, cha-ching!
Mining 4 done, cha-ching!
Mining 5 done, cha-ching!
Mining 6 done, cha-ching!
Mining 7 done, cha-ching!
Mining 8 done, cha-ching!
Mining 9 done, cha-ching!
Mining 10 done, cha-ching!
Mining 11 done, cha-ching!
Mining 12 done, cha-ching!
Mining 13 done, cha-ching!
Mining 14 done, cha-ching!
Mining 15 done, cha-ching!

What we want, with non-blocking communication is something more like this:

church$ ./bin
Mining 2 done, cha-ching!
Mining 3 done, cha-ching!
Mining 10 done, cha-ching!
Mining 13 done, cha-ching!
Mining 6 done, cha-ching!
Mining 8 done, cha-ching!
Mining 12 done, cha-ching!
Mining 15 done, cha-ching!
Mining 11 done, cha-ching!
Mining 7 done, cha-ching!
Mining 9 done, cha-ching!
Mining 1 done, cha-ching!
Mining 14 done, cha-ching!
Mining 0 done, cha-ching!
Mining 5 done, cha-ching!
Mining 4 done, cha-ching!

• Question 1: Why does blocking approach give you bursty responses? I.e. a bunch cache in at the same time.
• Question 2: What impact will allowing the children to cache out in any order have?
• TODO 1: Change the implementation to use non-blocking communication. Please note that there is already a nbpipe that has a non-blocking read end.
• Question 3: Is non-blocking faster? In what way? In what way isn't it faster?
• TODO 2: We want this all be very fast, so we want to avoid the copies of the pipe-based, message-passing implementation. Please use the shared memory (already set up in shmem for each child) to pass the data back to the parent and cash in.