NOTE: this is run on a "slightly modified" version from the previous assignment, due to:
- bug fixes
- changes to how offline works: the assignment will send a delayed response, but this implementation will ignore the message all together, since testing with too many nodes, too many runs will create too many threads, crashing the program
- remove all println and stuff
- add message count to result (doesn't affect the logic)
the reason why i choose my implementation:
- can't get the dependencies setup for the existing paxos implementations
- i don't like tencent (biased opinion). tho i can't set it up anyways
- don't want to write build scripts
- have no idea how to test code on there
- my code is already working pretty well
- i'm pretty good at automated testing
- MOST IMPORTANTLY: lazy
parameters that might affect runtime performance:
- number of nodes in the network:
- impact: A LOT
- e.g. 2 concurrent nodes, 10 runs
- 9 nodes:
30mssuccess,15msfail - 99 nodes:
600mssuccess,400msfail
- number of nodes propose concurrently:
- impact: not much, BUT it can skew the average rejected proposal time
- the more nodes propose concurrently, the higher average rejected proposal time is
- usually succeeded proposal takes longer, but if too many nodes propose at a time rejected proposals eventually pass the succeeded proposal (due to processing, maybe?)
- number of node failed:
- impact: A LOT
- timeout is set to 1 SECOND for each phase
- add at least 2-3 seconds to succeeded proposal & 1-2 seconds to rejected proposals
how to run:
# build
javac -d bin *.java
java -cp bin Test [args0] [args1] [args2] ([optional args3])
# args0 - default = 9 : number of nodes in the network
# args1 - default = 2 : number of nodes propose at the same time
# args2 - default = 10 : number of runs to try and average out, recommend 10 runs
# args3 - default = 0 : number of nodes failed (-> 0 = no failure mode)
# NOTE: i will not check for invalid input