Distributed Transaction Management, Fall 2002, 2-3 cu

(Hajautettujen transaktioiden hallinta, Syksy 2002, 2-3 ov)


General information




General information

If there are foreign students present, then the teaching will be given in English. The course material will be in English in any case.

The exercise sessions are on Wednesdays, this way you can make questions on Tuesdays, if there are some problems about the exercises. There are no separate exercise groups or sessions - we will do them at the start of the Wednesday sessions.

The course will include a coursework, which can be programming oriented or then not. It is possible to gain an extra credit by doing a larger coursework, which combines both theory and practice. In some ways, this is a fairly traditional computer science course. There are lectures (which may not always be like traditional lectures, though), and weekly exercises. There is also a courseworks, which is a programming project, and an exam.

You may take material to the exam.

The teaching for this course will run for 5 weeks. I will be in Helsinki on 30.10., which means that there will be no teaching on that particular day, and the first exercise session will surprisingly be on 29.10., which is a Tuesday. We can estimate the last teaching day to be 20.11.

As this course largely overlaps with the old Distributed System Implementation course, it is not so easy to give full to students who have done DSI earlier. There is some new material as well, so if any of those people wants to attend, it is possible, but needs to do an extended coursework or something like that.


  First lecture

Second lecture

Third lecture

Fourth lecture

Fifth lecture

Sixth lecture

Seventh lecture

Check out the distributed commit simulation applet!

Exercise information

Exercises for 29.10.2002

We will discuss these only initially on 29.10.
  1. What are "Primary Copy 2PL" and "Voting 2PL" as presented in the survey by Bernstein and Goodman in terms of lecture handouts?
  2. Is it possible for Chandy-Misra-Haas to detect a phantom deadlock, if strict 2PL is used and we the "probe" messages include the transaction id of the originator of the probe?

Exercises for 12.11.2002

  1. Assume that we want 2PC (two-phase commit) to be able to recover from the situation where only the coordinator crashes and the others time out, having voted yes. What, in the current formulation of 2PC termination protocol, is against this recovery?
  2. How can the recovery of the previous exercise be organised?
  3. Show that if the network may disconnect, then any distributed commit protocol may block.