Distributed Transaction Management, Autumn 2003 , Exercise 5
Due Tue 25.11.2003

1. Give a state diagram for a decentralised 3PC participant (ie, when
everyone sends everyone the messages, see for instance the simulation
applets nonblocking centralised protocol).

2. Could a transaction requesting locks only from one site cause a
global deadlock ? Can Chanhy-Mistra-Haas algorithm found this deadlock ?

Whoever is interested, may study the paper "Distributed Deadlock
Detection" by K. M. Chandy, J. Misra, and L. Haas (ACM Transactions on
Computer Systems, Vol 1, No 2, 1983, Available through ACM Digital
Library for those who have access to the system (www.acm.org).

3. Suppose the following execution of transactions:


T1         T2
startTxn
	   startTxn
read(x)
           write(y)
           write(z)
           commit
read(y)
commit

Explain what happens in forward validation and backward validation.

4. Suppose that locking is not 2PL (that is, you may now release locks
before taking new ones). Show how Chandy-Misra-Haas algorithm may
discoreve a phantom deadlock. (Easy)

5. Suppose you want to use timestamping for solving conflicts between a
write and a read, and locking to solve conflicts between writes. Under
which conditions does this seem like a better idea than just using
locks?