The questions for the reading on multi-core MapReduce:

1: Amdhal's Law (discussed in the first lecture) claims that the speedup that can be obtained with N cores running in paralel is at most N. Give an example from the Phoenix system benchmarking experiment that seemingly contradicts this claim and explain how the results relate to Amdhal's Law and why there is no contradiction.

2: Chris the sysadmin of COSI department suspects that Willy the Hacker got hold of the encrypted password file for COSI department user accounts and has mounted a "dictionary attack" (see definition in Wikipedia) to crack the passwords. To check in a hurry what passwords can be cracked using this attack Chris is about to run the attack himself using the MapReduce framework of the Phoenix system running on an 8 core CMP. He needs to set the unit size for the Map step. Chris knows that larger units impose less overhead but is not sure what size to choose for fastest results. Can you help Chris set the unit size?