Unit 6 : Distributed Deadlock Detection Distributed Systems
- 1. Unit 6 : Distributed Deadlock Detection Introduction, Preliminaries, Deadlock handling strategies in Distributed Systems, Issues in Deadlock Detection and Resolution, Control organizations for distributed deadlock detection, Centralized deadlock detection algorithms, Distributed deadlock detection algorithms, Avoidance and Prevention algorithms, Hierarchical deadlock detection algorithms
- 2. Introduction A distributed system is a network of sites that exchange information with each other by message passing. In DS, a process can request and release resources in any order, which may not be known prior and a process can request some resources while holding others. if the sequence of the allocation of resources to processes is not controlled in such environments, deadlock can occur. Several deadlock detection techniques based on various control organizations are described in this chapter.
- 3. Preliminaries The problems of deadlocks has been generally studied in DS under following model: The system have only reusable resources. Processes are allowed only exclusive access to resources. There is only one copy of each resource. A process can be in 2 states: running or blocked. In running (active) state, a process has all the needed resources and is either executing or is ready for execution. In blocked state, a process is waiting to acquire some resources.
- 4. 2 types of deadlocks: Resource Deadlocks: Process can simultaneously wait for several resources and cannot proceed until they have acquired all those resources. A set of processes is resource-deadlocked if each process in the set requests resources held by another process in the set and must receive all of the requested resources before it can become unblocked. Communication Deadlocks: In this, processes, communicate with other processes among a set of processes. A waiting process can unblock on receiving a communication from any one of these processes. Preliminaries (contd.)
- 5. A graph-theoretic model: The state of a process-resource interaction is shown by resource-allocation graph. The node of this graph are processes and resources of a system and edges of a graph depict assignment or pending request. A resource assignment is represented by an assignment edge directed from the node of an assigned resource to the node of the assigned process A system is deadlock if its resource allocation graph contains a directed cycle or a knot. Preliminaries (contd.)
- 6. In DS, the system state can be modeled or represented by a directed graph, called a wait- for-graph(WFG). In DDBS, users access the data objects of the database by executing transactions. A transaction can be viewed as process that performs a sequence of reads and writes on the data objects. In DDBS, WFG is referred as transaction- wait-for graph(TWF graph) Preliminaries (contd.)
- 7. Deadlock handling strategies in Distributed Systems Deadlock handling is complicated to implement in DS because no one site has accurate knowledge of the current state of the system so involve finite and unpredictable delay. 3 deadlock handling strategies are: Deadlock prevention Deadlock avoidance Deadlock detection
- 8. Deadlock prevention It is commonly achieved by either having a process acquire all the needed resources simultaneously before it begins execution or by preempting a process that holds the needed resources. In this, a process requests(or releases) a remote resource by sending request message to the site where the resource is located. Some drawbacks of this are listed below: It is inefficient as it decreases the system concurrency. A set of resources can become deadlocked in the resource acquiring phase. In many systems, future resource requirements are unpredictable.(i.e. not known a priori).
- 9. A resource is granted to a process if the resulting global system (including all the processes and resources of DS) is safe. Problems causing deadlock avoidance are: 1. Every site has to maintain information on global state of the system, which translates into huge storage requirements and extensive communication costs 2. The process of checking for a safe global state must be mutually exclusive. This restriction will severely limit the concurrency and throughput of the system. 3. Due to large number of processes and resources, it will be computationally expensive to check for Deadlock Avoidance
- 10. It requires an examination of the status of process-resource interactions for the presence of cyclical wait. Deadlock detection in DS has 2 favorable conditions: 1.Once a cycle is formed in WFG, it persists until it is detected and broken. 2.Cycle detection can proceed concurrently with the normal activities of a system. Deadlock Detection
- 11. Issues in Deadlock Detection and Resolution Deadlock detection has 2 basic issues: Detection of existing deadlocks Resolution of detected deadlocks. Detection: 2 issues: 1) maintenance of WFG 2) search of the WFG for the presence of cycles. Depending upon manner in which WFG information is maintained , the search for cycles is carried out. There are centralized, distributed and hierarchical algorithms for deadlock detection.
- 12. A correct deadlock detection algorithm must satisfy following conditions: Progress-No Undetected deadlocks. The algo. must detect all existing deadlocks in finite time. Once a deadlock has occurred the deadlock detection activity should continuously progress until the deadlock is detected. Safety-No False deadlocks. The algo. should not report deadlocks which are non-existent Called as Phantom deadlocks. Resolution: It involves breaking existing wait-for dependencies in the system WFG to resolve the deadlock. Several deadlock detection algo. Propogate information regarding wait-for dependencies along the edges of the wait-for-graph. So when wait-for dependency is broken, the corresponding information should be immediately cleaned from the system. If it is not cleaned properly in a timely manner, it may result in detection of Phantom deadlocks. Issues in Deadlock Detection and Resolution
- 13. Control organizations for distributed deadlock detection Centralized control: In this a designated (control) site has responsibility of constructing the global WFG and searching it for cycles. These are conceptually simple and easy to implement. Here the control site has complete information about the deadlock cycle and it can thus optimally resolve the deadlock. But it has single point of failure: communication links near the control site are likely to be congested because the control sites receives WFG information from all other sites. Also, message traffic generated due to deadlock detection activity is independent of the rate of deadlock formation and structure of the deadlock cycles.
- 14. Distributed control: In this the responsibility of detecting global deadlock is shared equally among all sites. The global state graph is spread over many sites and several sites participate in the detection of global cycles. In this, control is not vulnerable to a single point of failure and no site is swamped with deadlock detection activity. These algorithms are difficult to the lack of globally shared memory. Also in this several sites may initiate deadlock detection for the same deadlock. Proof of correctness is difficult for these algorithms. Control organizations for distributed deadlock detection
- 15. Hierarchical control: In this sites are arranged in a hierarchical fashion and site detects deadlocks involving only its descendent sites. These require special care while arranging all sites the sites in the hierarchy. For efficiency, most deadlocks should be localized to as few clusters as possible. Control organizations for distributed deadlock detection
- 16. Centralized deadlock detection algorithms The completely centralized algorithm: Designated site is called control site Maintains the WFG of the entire system and checks for existence of deadlock cycles. All sites request or release resources by sending request resource or release resource messages. Accordingly WFG will be updated Simple and easy to implement Highly inefficient because all resource acquisition and release request must go through the control site, even when resource is local. It results in large delays in responding to user requests, large communication overhead and congestion of communication links near the control site. Reliability is poor because if control site fails, the entire system come to halt because all the status information resides at the control site.
- 17. The completely centralized algorithm (contd.): Due to inherent communication delays and lack of perfectly synchronised clocks, the designated site may get an inconsistent view of the system and detect false deadlocks. T1 (S3) T2(S4) Lock R1 Lock R1 Unlock R1 Unlock R1 Lock R2 Lock R2 Unlock R2 Unlock R2 Centralized deadlock detection algorithms S1 R1 S2 R2 Designated Site T2 T1 T1 T2 T1-> T2 -> T1
- 18. The Ho-Ramamoorthy Algorithm: Two-phase algorithm One-phase algorithm These algorithms collect two consecutive status reports or keep two status tables at each site gets a consistent view of the system. Centralized deadlock detection algorithms
- 19. The Ho-Ramamoorthy Algorithm: Two-phase algorithm Every site maintains a status table that contains all the processes initiated at that site Includes all resources which are locked and in waited condition Periodically WFG is constructed at designated site If no cycle -> system is free from deadlock If cycle -> request for status table from all sites & again construct WFG using only those transactions which are common to both reports. If same cycle is detected, the system is declared deadlock. Centralized deadlock detection algorithms
- 20. The Ho-Ramamoorthy Algorithm: One-phase algorithm Requires only one status report from each site. Each site maintains 2 status tables: resource status table : keeps track of the transactions that have locked or are waiting for resources stored at that site. process status table : keeps track of the resources locked by or waited for by all the transactions at that site. It does not detect false deadlocks because it eliminates the inconsistency in state information by using only the information that is common to both tables. Centralized deadlock detection algorithms
- 21. Distributed deadlock detection algorithms Algorithms can be divided into 4 classes: Path-pushing Edge-chasing Diffusion computation Global state detection
- 22. Distributed deadlock detection algorithms Path-pushing Algorithms : Wait for dependency information of the global WFG is disseminated in the form of paths (i.e. sequence of wait-for dependency edges) This algorithm sends n(n-1)/2 messages to detect a deadlock involving n sites. Size of message is O(n). Delay detecting deadlock is O(n)
- 23. Distributed deadlock detection algorithms Edge-chasing Algorithms : Special messages called probes are circulated along the edge of the WFG to detect a cycle. When a blocked process receives a probe, it propagates the probe along its outgoing edge in WFG. A process declares a deadlock when it receives a probe initiated by it. Here probes are of fixed size
- 25. Distributed deadlock detection algorithms Diffusion Computation Based Algorithms : It make use of Echo algorithm to detect deadlocks. For this, process sends out query message along all the outgoing edges in WFG. Queries are discarded by running process and are echoed back by blocked processes in the following way: When a blocked process receives first query message for a particular deadlock detection initiation, it does not send a reply message until it has received a reply message for every query it has sent. For all subsequent queries, it immediately sends back a reply message. The initiator of a deadlock detection detects a deadlock when it receives a reply message for every query it sent.
- 26. Distributed deadlock detection algorithms Global State Detection Based Algorithms : It has following facts: A consistent snapshot of DS can be obtained without freezing the underlying computation. A consistent snapshot may not represent the system state at any moment in the time, but if a stable property holds in the system before the snapshot collection is initiated, this property will still hold in the snapshot.
- 27. Hierarchical deadlock detection algorithms Sites are (logically) arranged in the hierarchical fashion. A site is responsible for detecting deadlocks involving only its children sites These algorithms take advantage of access patterns that are localized to a cluster of sites to optimize performance.
- 28. Hierarchical deadlock detection algorithms The Menasce-Muntz algorithm: All the controllers are arranged in tree fashion The controllers at bottommost level (leaf controllers) manage resources, and other (non-leaf controllers) responsible for deadlock detection Whenever a change occurs in a controller’s TWF graph due to resource allocation, wait or release is propagated to its parent controller The parent controller makes changes in its TWF graph, searches for cycles, and propagates the changes upword, if necessary A non-leaf controllers can receive up-to-date information concerning the TWF graph of its children continuously
- 29. Hierarchical deadlock detection algorithms The Ho- Ramamoorthy algorithm: Sites are grouped into several disjoint clusters. A site is chosen as central control site which dynamically chooses a control site for each cluster A central control site request from every control site their intercluster transaction status information and wait for relations