Principles Of Distributed Database Systems Exercise Solutions ^hot^ Jun 2026

Suppose we have a distributed database system with three nodes, each storing a different fragment of a large database. We want to process a query that retrieves all customers who have placed an order for a specific product.

I can help walk through the calculations or the theoretical, step-by-step solutions. What Is a Distributed Database? | Oracle Saudi Arabia

Lock managers are distributed across all sites. Exercises test your ability to trace lock requests and identify distributed deadlocks. Distributed Deadlock Detection

Does the user feel like they are using a single database? Suppose we have a distributed database system with

Is there a global deadlock? How to detect?

Employee=EmployeeVF1⋈EmpIDEmployeeVF2Employee equals Employee sub cap V cap F 1 end-sub ⋈ sub EmpID end-sub Employee sub cap V cap F 2 end-sub Horizontal: If a tuple appears in fragment Ricap R sub i , it must not appear in any other fragment Rjcap R sub j ). Since the predicates are mutually exclusive ( ), the fragments are strictly disjoint:

Consistency condition: read quorum + write quorum > N (number of replicas). Here 2+4=6 >5 ✔. If R=2, W=4, then concurrent writes must intersect on at least one replica (4+4>5? No – two write quorums intersect if 4+4>5 → 8>5 ✔). Failure tolerance: To block a write, you need to knock out 5-4+1=2 replicas (since write quorum=4, failing 2 leaves 3 available – not enough for write). To block a read, you need to knock out 5-2+1=4 replicas (failing 4 leaves 1 < read quorum). So failure tolerance = min(2,4)-1? Let’s compute correctly: Write quorum=4 → max failures for availability = N-W =1. Read quorum=2 → max failures for availability = N-R=3. System can survive at most 1 failure and still perform writes. What Is a Distributed Database

Profile update → eager replication (strong consistency). Like counter → lazy replication (eventual consistency).

How can a user query a table without knowing it is split across servers in New York and London?

We can allocate these fragments to nodes as follows: Distributed Deadlock Detection Does the user feel like

Check the graph for cycles. If the graph is a Directed Acyclic Graph (DAG), the schedule is conflict-serializable; if a cycle exists, the schedule must be rejected.

write_TS(D)=5, read_TS(D)=5. TS(T2)=20 > both → allow. Update write_TS(D)=20.

Compute R ⋉ S: project R.B values that appear in S.B. Assume 80% of R.B in S.B → 800 tuples. Ship R' (800 tuples) to Site2 to join with S → result RS of 400 tuples (estimated). Ship RS to Site1 to join with T: 400 ⨝ 2000 = 400 tuples final (assuming foreign key). Cost: transfer R' (800) + transfer RS (400) = 1200.