Written Assignment Unit 2
University of the People
Department of Computer Science
CS 3306-01 - DATABASE 2 - AY2025-T2
Dr. Brian Stevens (Instructor)
November 21st, 2024
Deadlock Recovery: Actions for Breaking Deadlocked Transactions in Database
Systems
When a detection algorithm determines that a deadlock exists, the system must recover from the
deadlock. The most common solution is to roll back one or more transactions to break the
deadlock. List the three actions that need to be taken using Figure 16.19 to discuss.
Figure 16.19, as depicted in the textbook, is provided below:
From the wait-for graph above, we can observe that the system comprises four transactions: T25, T26,
T27, and T28. T26, T27, and T28 are deadlocked transactions because they form a cyclic connection
within the graph.
According to Silberschatz et Al (2001), to resolve a transaction deadlock in a database, some or all of the
deadlocked transactions must be rolled back, partially or totally. This process involves three essential
actions that need to be taken, which are:
Selection of victim
Rollback
Starvation
Each action is discussed below, using the wait-for graph as a reference.
Selection of Victim
To resolve a deadlock, the system must identify and terminate (or roll back) one or more transactions to
break the cyclic dependency. The aim of this action is to choose a victim that incurs the least cost upon
rollback. We do so based on the following criteria, as elaborated by Silberschatz et Al (2001):
The duration the transaction has been running and the estimated time remaining to complete its task.
The number of data items the transaction has already accessed.
The additional data items are required for the transaction to finish.
The number of transactions that will be affected by the rollback.
To break the cycle illustrated in Figure 16.19, the system identifies that T26 is holding resources needed
by T27, T27 is holding resources required by T28, and T28 is waiting for resources held by T26. The
system then evaluates the rollback cost for each transaction. If T28 has the lowest rollback cost, it will
be chosen as the victim and terminated. Rolling back T28 breaks the cycle, allowing the remaining
transactions to proceed.
Rollback
After selecting a victim transaction, such as T28 from the wait-for graph in Figure 16.19, the next step is
to roll back the transaction to release the resources it holds, effectively breaking the deadlock cycle.
Rolling back involves undoing all or part of the transaction to restore the system to a consistent state. As
elaborated by Silberschatz et Al (2001), there are two main types of rollbacks that can be performed:
Partial Rollback:
Instead of aborting the entire transaction, only part of its operations is undone. The transaction is rolled
back to a specific point where the deadlock can be resolved, after which it is restarted or continues from
that point.
Total Rollback:
The transaction is fully aborted, and all of its changes to the database are undone. This ensures that all
resources held by the transaction are released, allowing other transactions to proceed.
This means that In the wait-for graph if T28 is selected as the victim, it can either be completely rolled
back (terminated) or partially rolled back to a point where it no longer holds the resources required by
T26. By releasing the resources held by T28, T26 and T27 can resume their execution, effectively
breaking the deadlock cycle.
Starvation
Starvation happens when a transaction is repeatedly chosen as a victim and rolled back during deadlock
resolution, preventing it from ever completing its execution (Silberschatz et Al., 2001). To avoid
starvation, the system must guarantee that every transaction eventually makes progress and is completed
(Silberschatz et Al., 2001). This can be achieved by implementing strategies that ensure fairness in
victim selection.
In the given wait-for graph, if T28 is continuously selected as the victim due to its lowest rollback cost,
it may never finish its execution, leading to starvation. To prevent this, the system must implement
mechanisms to ensure that T28 is eventually allowed to complete.
In conclusion, these are the three actions that are taken when a system comes across deadlocks and
needs to take care of them using deadlock detection algorithms and deadlock recovery mechanisms.
Reference:
Silberschatz, A., Korth, H.F., & Sudarshan, S. (2001). Database System Concepts (4th ed.). New York,
NY: McGraw-Hill. Available at Database System Concepts 4th Edition By Silberschatz-Korth-
Sudarshan.pdf)
