RECOVERY:

Consider a transaction that updates 1000 pages in some relation.  There are
only 500 pages of buffer available to this transaction.  Explain what would
happen if a "steal" policy were not in force for buffer aanagement.

Suppose I can partition a database into chunks such that each transaction
only accesses data within one chunk.  I now build a checkpoint scheme where I
quiesce one chunk at a time and checkpoint it.  Is the resulting checkpoint
fuzzy -- why or why not?  
Do the same problem as above where each transaction only updates data within
one chunk, but may read data from other chunks.

Page P is written to disk at time t=0.  Page Q is written to disk at a later
time t >0.  Is it possible for pageLSN(Q) < pageLSN(P)?  Why or why not?

Give an example of eager non-transactional replication.

Name one advantage/disadvantage each of eager versus lazy replication.