Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Distributed Systems-Unit 2

9,727 views

Published on

Distributed Mutual Exclusion: Classification of distributed mutual exclusion, requirement of
mutual exclusion theorem, Token based and non token based algorithms, performance metric for
distributed mutual exclusion algorithms.
Distributed Deadlock Detection: system model, resource Vs communication deadlocks,
deadlock prevention, avoidance, detection & resolution, centralized dead lock detection,
distributed dead lock detection, path pushing algorithms, edge chasing algorithms.

Published in: Engineering
  • Be the first to comment

Distributed Systems-Unit 2

  1. 1. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA UNIT -2 Distributed Mutual Exclusion The mutual exclusion (ME) problem frequently arises when concurrent access to shared resources by several sites is involved (e.g., directory management where updates and reads to a directory must be done atomically to ensure correctness.) Mutual exclusion is a fundamental issue in the design of distributed systems, and an efficient and robust technique for mutual exclusion is essential. Because of resource distribution, transmission delays, and lack of global information, mutual exclusion algorithms developed for single-computer systems are not applicable in a distributed environment. Basic approaches to distributed mutual exclusion: Algorithms developed for distributed systems can be grouped into two classes: 1. Non-token-based (assertion based) approach: a site enters its critical section (CS) when an assertion defined on its local variables becomes true, which can be established through two or more rounds of message exchanges. 2. Token-based approach: a site enters its CS if it possesses the single token that is shared among the sites. System model 1. At any instant, a site may have several requests for CS. The site queues them up and serves them one at a time. 2. A site can be in one of three states:  Requesting CS -- the site is blocked and cannot make further requests for CS  Executing CS -- the site is in its critical section  Idle -- the site is neither requesting nor executing CS Objectives of Mutual Exclusion Algorithms 1. Guarantee mutual exclusion (required) 2. Freedom from deadlocks (desirable) 3. Freedom from starvation -- every requesting site should get to enter CS in a finite time (desirable) 4. Fairness -- requests should be executed in the order of arrivals, which would be based on logical clocks (desirable) 5. Fault tolerance -- failure in the distributed system will be recognized and therefore not cause any unduly prolonged disruptions (desirable)
  2. 2. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Performance Metric There are 4 metrics to Measuring the performance of ME algorithms 1. Number of messages necessary per CS invocation 2. Synchronization delay, the time required after a site leaves the CS and before the next site enters the CS. Message exchanges are required. 3. Response time, the time interval a request waits for its CS execution to be over after its request messages have been sent out 4. System throughput, the rate at which the system executes requests for CS, defined by 1/(sd+E), where sd is the synchronization delay, and E is the average critical section execution time. Notes on performance measurements 1. Algorithm performance may depend on the loading conditions of the system. Two conditions are of particular interest: low load (ll) where there is seldom more than one request for CS simultaneously in the system and high load (hl) where there is always a pending request for CS at a site. 2. Algorithms are often studied for best case and worst case performances. Often the best and worst cases coincide with low and high loads (e.g., the best and worst values of the response time are achieved when the load is low and high, respectively.) Synchronization delay time CS execution time time FIGURE 6.1 Synchronization delay The site enters the CS Its request message sent out CS request arrives The site exits the CS Response time FIGURE 6.2 Response time Last site exits CS Next site enters CS
  3. 3. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA NON-TOKEN-BASED ALGORITHMS In non-token-based mutual exclusion algorithms, a site communicates with a set of other sites to arbitrate who should execute the CS next. Each site ‘Si’ maintains a request set ‘Ri’ that contains ids of all those sites from which it must acquire permission before entering the CS. A site can proceed to its CS only after it has acquired permission from all those sites in its request set. Messages are time stamped using Lamport’s logical clock scheme. Messages with the same virtual timestamp are resolved by the ids of the message sending sites. Smaller timestamp requests have priority over larger timestamp requests. Lamport’s Algorithm Lamport gave the following distributed mutual exclusion algorithm as an illustration of his scheme of logical clocks. In this algorithm, every site Si, 1<i<N, maintains a request set Ri = {S1, S2,......, SN}, and a queue called request queue ‘RQi’ that contains mutual exclusion requests it has received and not yet released in order of their timestamps. The algorithm requires messages to be delivered in the FIFO order between every pair of sites. The Algorithm 1. Requesting the critical section  When a site Si wants to enter the CS, it sends a REQUEST (tSi, i) message to all the sites in its request set Ri and places the request on Qi. Its placement in the queue is based on its timestamp (tSi,i).  When a site Sj receives the REQUEST (tSi, i) message from site Si, it returns a timestamped REPLY message to Si, and place site Si’s request on request queue RQi. 2. Executing the critical section  Site Si enters the CS when the following two conditions hold: o [L1]: Si receives a message with timestamp larger than (tSi, i) from all other sites. o [L2]: Si’s request is at the top of RQi. 3. Releasing the critical section  Site Si, upon exiting the CS, removes its request from the top of RQi and sends a time stamped RELEASE message to all sites in Ri.  When a site Sj receives a RELEASE message from site Si, it removes Si’s request from RQj.
  4. 4. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA  When a site removes a request from its request queue, its own request may come at the top of the queue, enabling it to enter the CS.
  5. 5. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA The Lamport’s algorithm executes CS requests in the increasing order of timestamps. Space-time diagrams of example to illustrate how the Lamport’s algorithm works. The algorithm’s correctness can be proved by contradiction. Illustration of Lamport’s Algorithm Performance Lamport’s algorithm requires 3(N-1) messages per CS invocation: (N-1) REQUEST, (N-1) REPLY, and (N-1) RELEASE messages. Synchronization delay in the algorithm is T, the average message delay. An Optimization Lamport’s algorithm can be optimized to require between 3(N-1) and 2(N-1) messages per CS invocation by suppressing REPLY messages in the following situation: Site Sj receives a REQUEST message from site Si after it has sent its own REQUEST message with timestamp higher than the timestamp of site Si’s request. This is possible since Sj’s REQUEST message would have satisfied condition [L1] in Si’s decision for entering the CS. The Ricart - Agrawala Algorithm The Ricart-Agrawala algorithm is an optimization of Lamport’s algorithm that dispenses with RELEASE messages by cleverly merging them with REPLY messages. Again in this algorithm, every site Si, 1 < I < N, maintains request set Ri = {S1, S2, S3, ......, SN}, The Algorithm 1. Requesting the critical section  When a site Si wants to enter the CS, it sends a REQUEST message to all the sites in its request set Ri.  When a site Sj receives the REQUEST message from site Si, it sends a REPLY message to Si if site Sj is neither requesting nor executing the CS or if Sj is requesting and Si’s request’s timestamp is smaller than site Sj’s own request’s timestamp. The request is deferred otherwise. 2. Executing the critical section Site Si enters the CS after it has received REPLY messages from all sites in its request set Ri.
  6. 6. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA 3. Releasing the critical section When site Si exits the CS, it sends REPLY messages to all sites it has deferred REPLY to. Illustration of Ricart-Agrawala Algorithm
  7. 7. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Space-time diagrams of example to illustrate how the Ricart-Agrawala algorithm works. The correctness of the algorithm can also be proved by contradiction. Performance: The Ricart-Agrawala algorithm requires 2(N-1) messages per CS invocation: (N-1) REQUEST and (N-1) REPLY messages. Synchronization delay in the algorithm is also T, the average message delay, as in the Lamport’s algorithm. Maekawa’s Algorithm In Maekawa’s algorithm, a site requests permission, only from a subset of sites instead of all sites. To prevent two sites to obtain all the necessary permissions for CS, these subsets of sites must have overlaps. Specifically, the request sets are chosen so that I j, 1 ≤ i , j ≤ N  RiRj ≠ . Each site receiving request messages serves as a mediator that ensures only one site under its watch can be given permission for CS at a time. Construction of Request Sets: The request sets for sites in Maekawa’s algorithm are constructed to satisfy the following conditions: M1: : ,1 , :: i ji j i j i j N R R        M2: :1 :: i ii i N S R    M3: :1 ::| |ii i N R K    M4: Any site jS is contained in K number of iR s, 1 ,i j N  , where N and K are related by the equation: ( 1) 1N K K   . This relation yields | |iR N . The Algorithm 1. Requesting the critical section.  When a site iS wants to enter the CS, it sends a REQUEST(i) message to all the sites in its request set iR .  When a site jS receives the REQUEST(i) message from site iS , it sends a REPLY(j) message to iS provided it hasn’t sent a REPLY message to a site from the time it received the last RELEASE message. Otherwise, it queues up the REQUEST for later consideration.
  8. 8. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA 2. Executing the critical section. Site iS enters the CS after it has received REPLY messages from all sites in its request set iR . 3. Releasing the critical section.  When site iS exits the CS, it sends RELEASE(i) messages to all sites in its request set iR .  When a site jS receives the RELEASE(i) message from site iS , it sends a REPLY(j) message to the next site waiting in the queue and deletes that entry from the queue. If the queue is empty, then the site updates its state to reflect that the site has not sent out any REPLY message. Performance: Maekawa’s algorithm requires 3 N messages per CS invocation: N REQUEST, N REPLY, and N RELEASE messages. Synchronization delay in the algorithm is 2T , twice as long as in both the Lamport’s and Ricart-Agrawala algorithms. In addition, Maekawa’s algorithm is deadlock-prone, and deadlock handling will require additional messages. Handling Deadlocks Maekawa’s algorithm handles deadlocks by requiring a site failing to lock all needed sites to yield a lock if the timestamp of its request is larger than the timestamp of some other request waiting for the same lock. A site suspects a deadlock whenever a higher priority request finds that a lower priority request has already locked the site (priority is based on timestamp), thus initiates message exchanges to resolve it. Messages required for deadlock handling are: FAILED: a FAILED message from site iS to site jS indicates that iS cannot grant jS ‘s request because it has currently granted permission to a site with a higher priority request. INQUIRE: an INQUIRE message from iS to jS indicates that iS would like to find out from jS if it has succeeded in locking all sites in its request set. YIELD: a YIELD message from site iS to jS indicates that iS is returning the permission to jS in order to yield to a higher priority request at iS .
  9. 9. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Deadlock handling procedure. 1. When a REQUEST( ,ts i) from site iS blocks at site jS because jS has currently granted permission to site kS , then jS sends a FAILED(j) message to iS if iS ’s request has lower priority. Otherwise, jS sends an INQUIRE(j) message to site kS . 2. In response to an INQUIRE(j) message from site jS , site kS sends a YIELD(k) message to jS , provided kS has received a FAILED message from a site in its request set or it sent a YIELD message to any of these sites but has not yet received a new GRANT from it. 3. In response to a YIELD(k) message from site kS , site jS assumes it has been released by kS , places the request of kS at the appropriate location in the request queue, and sends a GRANT(j) message to the top request’s site in the queue. These extra messages may be exchanged even though there is no deadlock. The maximum number of messages required per CS invocation with deadlock handling has been determined to be 5 N .
  10. 10. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA TOKEN-BASED ALGORITHMS In token-based mutual exclusion algorithms, a unique token is shared among all sites. A site can proceed to its CS if it possesses the token. The various algorithms differ in the way a site carries out the search for the token. Suzuki-Kasami’s Broadcast Algorithm The main design issues in this algorithm are: 1. distinguishing outdated REQUEST messages from current ones, 2. determining which site has an outstanding request for the CS. A REQUEST message of site jS has the form REQUEST(j,n), where n is a sequence number that indicates that site jS is requesting its nth CS execution. Each site iS maintains an array of integers [1.. ]iRN N where [ ]iRN j records the largest sequence number received so far in a REQUEST message from jS . A REQUEST(j,n) message received by iS is outdated if [ ]iRN j >n. Upon receiving REQUEST(j,n), site iS updates iRN by setting [ ]iRN j : = max( [ ], )iRN j n The token is associated with an array of integers [1.. ]LN N . After executing its CS, a site iS updates [ ] [ ]iLN i RN i to indicate that its request corresponding to sequence number [ ]iRN i has been executed. The token array LN enables site iS exiting its CS to determine if some other site has an outstanding request. All sites jS such that [ ] [ ] 1iRN j LN j  are waiting to execute their latest request of CS, and will be placed in the token queue Q. Site iS then sends the token to the site at the head of Q. The Algorithm 1. Requesting the critical section.  A site iS that wants to enter its CS increments its sequence number [ ]iRN i . If iS does not have the token, then it broadcast a REQUEST(i, sn) message to all sites, where sn is the updated value of [ ]iRN i .  When a site jS receives this message, it sets [ ]jRN i to max( [ ], )jRN i sn If jS has the idle token, then it compares [ ]jRN i and [ ]LN i , sends the token to iS if [ ] [ ] 1jRN i LN i  .
  11. 11. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA 2. Executing the critical section. Site iS executes the CS when it receives the token. 3. Releasing the critical section.  When site iS exits the CS, it sets [ ]LN i equal to [ ]iRN i .  For every site jS whose ID is not in the token queue Q, it appends its ID to Q if [ ] [ ] 1iRN j LN j  .  If Q is not empty, then it deletes the top site ID from Q and sends the token to the site indicated by the deleted ID. Performance: Suzuki-Kasami’s algorithm requires 0 to N messages per CS invocation. Synchronization delay in the algorithm is 0 or T . No message is needed and the synchronization delay is 0 if the site holds the token at the time of its request. Singhal’s Heuristic Algorithm Instead of broadcasting request messages for acquiring the token, Singhal’s algorithm has each site keep information about the state of other sites and use the information to select the sites that are likely to have the token. The objective of this effort is to reduce the number of messages required for the CS execution. It is called a heuristic algorithm because sites are heuristically selected for sending the request messages. In order to avoid the request being unanswered, a requesting site must send the request message to a site that either holds the token or is going to obtain the token in the near future. Hence one design requirement of this algorithm is that a site must select a subset of sites such that at least one of those sites is guaranteed to get the token in the near future. A site iS maintains two arrays [1.. ]iSV N and [1.. ]iSN N , to store information about other sites in the system. These arrays store the state and the highest known sequence number for each site, respectively. The token also contain two similar arrays, denoted by [1.. ]TSV N and [1.. ]TSN N . As in Suzuki-Kasami’s algorithm, sequence numbers are used to detect outdated requests. A site can be in one of the following states: R – requesting the CS E – executing the CS H – holding the idle token N – none of the above
  12. 12. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA The arrays are initialized as follows: For every site iS , 1..i N do [ ]:iSV j  N for ..j N i ; [ ]:iSV j  R for 1..1j i  ; [ ]: 0iSN j  for 1..i N 1[1]:S  := H For the token [ ]TSV j := N and [ ]: 0TSN j  for 1..j N It should be noted that the above initialization indicates site 1 hold the idle token initially, and for every two sites iS and jS , either [ ]iSV j = R or [ ]jSV i = R. This ensures that any two sites requesting for the token will always result in one sending a request to the other, hence no site is isolated and a site’s request will reach a site that is going to get the token in the near future. The Algorithm 1. Requesting the critical section.  A site iS that wants to enter its CS increments its sequence number [ ]iSN i and sets [ ]iSV i := R. If iS does not have the token, then it sends a REQUEST(i, sn) message to all sites jS for which [ ]iSV j = R, where sn is the updated value of [ ]iSN i .  When a site jS receives the REQUEST(i,sn) message, it discards the message if [ ]jSN i ≥sn which indicates that the message is outdated. Otherwise, it sets [ ]jSN i to sn and takes the following actions based on its own state: i. [ ]jSV j = N: Set [ ]jSV i := R. ii. [ ]jSV j = R: If [ ]jSV i ≠ R, then set [ ]jSV i := R and send a REQUEST(j, [ ]jSN j ) message to iS , else do nothing. iii. [ ]jSV j = E: Set [ ]jSV i := R. iv. [ ]jSV j = H: Set [ ]jSV i := R, [ ]TSV i := R, [ ]TSN i :=sn, [ ]jSV j := N, and send the token to iS .
  13. 13. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA 2. Executing the critical section. When Site iS receives the token, it first sets [ ]iSV i := E, then executes the CS. 3. Releasing the critical section.  When site iS exits the CS, it sets [ ]iSV i := N and [ ]TSV i := N, and updates its local and token vectors as follows: For all jS , j=1 to N do if [ ] [ ]iSN j TSN j then [ ]: [ ]; [ ]: [ ]i iTSV j SV j TSN j SN j  (* update token information from local information *) else [ ]: [ ]; [ ]: [ ]i iSV j TSV j SN j TSN j  (* update local information from token information *)  If ( :: [ ]ij SV j = N), then set [ ]iSV i := H, else send the token to a site jS such that [ ]iSV j = R. (Fairness of algorithm depends on policy used in selecting the site for receiving the token next.) Performance: Singhal’s algorithm allows a site to enter CS without communicating with every site in the system. In low to moderate loads, the message traffic is N/2 because a site needs to send REQUEST to approximately half the sites on average. At high loads, however, when every site is requesting for CS, then most entries in every site’s SV array have the value R, hence the message traffic increases to N. Synchronization delay in the algorithm isT . Raymond’s Tree-Based Algorithm In Raymond’s tree-based distributed mutual exclusion algorithm, sites are logically arranged as a directed tree rooted at the site holding the token. Every site maintains a local variable holder that points to its parent site in the tree structure. At the root site, holder points to itself. Every site also keeps a FIFO queue, called request_q, which stores the requests of those sites that have sent a request to this site, but have not yet been sent the token.
  14. 14. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA The Algorithm 1. Requesting the critical section.  A site iS that wants to enter its CS sends a REQUEST message to its parent site (indicated in the holder variable), provided it does not hold the token and its request_q is empty. It then adds its request to its request_q.  When a site on the path receives the REQUEST message, it places the REQUEST in the request_q and sends a REQUEST message to its parent along the directed path to the root, provided it has not sent out a REQUEST message for a previously received REQUEST on its request_q.  When the root site receives a REQUEST message, it sends the token to the site from which it received the REQUEST message, and sets its holder variable to point to that site.  When a site receives the token, it deletes the top entry from its request_q, sends the token to the site indicated in this entry, and sets its holder variable to point to that site. If the request_q is nonempty at this point, then the site sends a REQUEST message to its parent site as indicated in the holder variable. 2. Executing the critical section. When a site receives the token and the deleted top entry from its request_q is its own request entry, it sets the holder variable to point to itself and executes the CS. 3. Releasing the critical section.  When a site exits the CS, it checks its request_q. If its request_q is nonempty, it deletes the top entry from its request_q, sends the token to that site, and sets its holder variable to point to that site.  If the request_q is nonempty at this point, then the site sends a REQUEST message to the site which is pointed at by the holder variable. Example to illustrate how the Raymond’s algorithm works. Illustration of Raymond’s Algorithm
  15. 15. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA
  16. 16. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA The correctness of the algorithm in mutual exclusion enforcement is obvious since only one token is used. The algorithm is free from deadlock because the acyclic nature of the structure eliminates the possibility of circular wait among requesting sites. The algorithm is free from starvation can be rationalized based on two facts: (1) a site serves requests from its request_q in the FCFS order, and (2) every site has a path leading to the site that has the token. Performance: The average performance of Raymond’s algorithm (log )O N messages per CS invocation because the average distance between any two nodes in a tree with N nodes is (log )O N . Synchronization delay in the algorithm is ( log )/ 2T N . A Comparison of Performance (ll = light load, hl = high load) Non-Token Resp. time (ll) Sync Delay Messages (ll) Messages (hl) Lamport 2T E T 3( 1)N  3( 1)N  Ricart- Agrawala 2T E T 2( 1)N  2( 1)N  Maekawa 2T E 2T 3 N 5 N Token Resp. time (ll) Sync Delay Messages (ll) Messages (hl) Suzuki-Kasami 2T E T N N Singhal 2T E T / 2N N Raymond ( log )T N E ( log )/ 2T N log N 4 Universal Performance Bounds Metrics Bound Explanation Minimum synchronization delay T At least one message has to be transferred before the next site can receive the notification of permission to enter the CS Maximum throughput 1/( )T E The minimum time interval between two successive CS execution is T E Minimum response time 2T E One round trip message delay to acquire permission or the token plus time to execute the CS Maximum average response time ( )N T E This occurs at heavy loads (when all sites always have pending requests). Note that system executes requests at a rate 1/( )r T E  . From Little’s Law, response time = / ( )N r N T E  .
  17. 17. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Deadlock in Centralized System A deadlock is a condition where a process cannot proceed because it needs to obtain a resource held by another process and it itself is holding a resource that the other process needs. We can consider two types of deadlock: 1. Communication deadlock occurs when process A is trying to send a message to process B, which is trying to send a message to process C which is trying to send a message to A. 2. A resource deadlock occurs when processes are trying to get exclusive access to devices, files, locks, servers, or other resources. We will not differentiate between these types of deadlock since we can consider communication channels to be resources without loss of generality. Four conditions have to be met for deadlock to be present: 1. Mutual exclusion. A resource can be held by at most one process. 2. Hold and wait. Processes that already hold resources can wait for another resource. 3. Non-preemption. A resource, once granted, cannot be taken away from a process. 4. Circular wait. Two or more processes are waiting for resources geld by one of the other processes. We can represent resource allocation as a graph where: P ←R means a resource R is currently held by a process P. P→R means that a process P wants to gain exclusive access to resource R. Deadlock exists when a resource allocation graph has a cycle. Deadlocks in distributed systems Deadlocks in distributed systems are similar to deadlocks in centralized systems. In centralized systems, we have one operating system that can oversee resource allocation and know whether deadlocks are (or will be) present. With distributed processes and resources it becomes harder to detect, avoid, and prevent deadlocks.
  18. 18. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Several strategies can be used to handle deadlocks:  Ignore: we can ignore the problem. This is one of the most popular solutions.  Detect: we can allow deadlocks to occur, then detect that we have a deadlock in the system, and then deal with the deadlock  Prevent: we can place constraints on resource allocation to make deadlocks impossible  Avoid: we can choose resource allocation carefully and make deadlocks impossible. Deadlock avoidance is never used (either in distributed or centralized systems). The problem with deadlock avoidance is that the algorithm will need to know resource usage requirements in advance so as to schedule them properly. Deadlock detection General methods for preventing or avoiding deadlocks can be difficult to find. Detecting a deadlock condition is generally easier. When a deadlock is detected, it has to be broken. This is traditionally done by killing one or more processes that contribute to the deadlock. Unfortunately, this can lead to annoyed users. When a deadlock is detected in a system that is based on atomic transactions, it is resolved by aborting one or more transactions. But transactions have been designed to withstand being aborted. When a transaction is aborted due to deadlock:  system is restored to the state it had before the transaction began  transaction can start again  hopefully, the resource allocation/utilization will be different now so the transaction can succeed Consequences of killing a process in a transactional system are less severe. Centralized deadlock The centralized algorithm attempts to imitate the non-distributed algorithm by using a centralized coordinator. Each machine is responsible for maintaining its own processes and resources. The coordinator maintains the resource utilization graph for the entire system. To accomplish this, the individual sub graphs need to be propagated to the coordinator. This can be done by sending a message each time an arc is added or deleted. If optimization is needed (reduce messages) then a list of added or deleted arcs can be sent periodically.
  19. 19. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Suppose machine A has a process P0 which holds resource S and wants resource R. Resource R is held by P1. This local graph is maintained on machine A. Suppose that another machine B, has a process P2, which is holding resource T and wants resource S. Both of these machines send their graphs to the coordinator, which maintains the union (overall graph). The coordinator sees no cycles. Therefore there are no deadlocks. If a cycle was found (hence a deadlock), the coordinator would have to make a decision on which machine to notify for killing a process to break the deadlock. Centralized deadlock detection We use a centralized deadlock detection algorithm and try to imitate the non‐distributed algorithm.  Each machine maintains the resource graph for its own processes and resources.  A centralized coordinator maintains the resource graph for the entire system.  When the coordinator detects a cycle, it kills off one process to break the deadlock.  In updating the coordinator’s graph, messages have to be passed.  Method 1: Whenever an arc is added or deleted from the resource graph, a message has to be sent to the coordinator.  Method 2: Periodically, every process can send a list of arcs added and deleted since previous update.  Method 3: Coordinator asks for information when it needs it.
  20. 20. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Centralized Deadlock‐Detection Algorithms The Ho‐Ramamoorthy Algorithms 1. The Two‐Phase Algorithm 2. The One‐phase Algorithm Ho‐Ramamoorthy 2‐phase Algorithm 1. Each site maintains a status table of all processes initiated at that site: includes all resources locked & all resources being waited on. 2. Controller requests (periodically) the status table from each site. 3. Controller then constructs wait for graph (WFG) from these tables, searches for cycle(s). 4. If no cycles, no deadlocks. 5. Otherwise, (cycle exists): Request for state tables again. 6. Construct WFG based only on common transactions in the 2 tables. 7. If the same cycle is detected again, system is in deadlock. 8. Later proved: cycles in 2 consecutive reports need not result in a deadlock. Hence, this algorithm detects false deadlocks. Ho‐Ramamoorthy 1‐phase Algorithm 1. Each site maintains 2 status tables: resource status table and process status table. 2. Resource table: transactions that have locked or are waiting for resources. 3. Process table: resources locked by or waited on by transactions. 4. Controller periodically collects these tables from each site. 5. Constructs a WFG from transactions common to both the tables. 6. No cycle, no deadlocks. 7. A cycle means a deadlock. Distributed Deadlock‐Detection Algorithms A Path‐Pushing Algorithm 1. The site waits for deadlock‐related information from other sites 2. The site combines the received information with its local transaction wait for (TWF) graph to build an updated TWF graph 3. For all cycles ‘EX ‐> T1 ‐> T2 ‐> Ex’ which contains the node ‘Ex’, the site transmits them in string form ‘Ex, T1, T2, Ex’ to all other sites where a sub‐transaction of T2 is waiting to receive a message from the sub‐transaction of T2 at that site
  21. 21. LECTURE NOTES: DISTRIBUTED SYSTEM (ECS-701) MUKESH KUMAR DEPARTMENT OF INFORMATION TECHNOLOGY I.T.S ENGINEERING COLLEGE, GREATER NOIDA PLOT NO: 46, KNOWLEDGE PARK 3, GREATER NOIDA Edge‐Chasing Algorithm Chandy‐Misra‐Haas’s Algorithm: 1. A probe(i, j, k) is used by a deadlock detection process Pi. This probe is sent by the home site of Pj to Pk. 2. This probe message is circulated via the edges of the graph. Probe returning to Pi implies deadlock detection. 3. Terms used: a. Pj is dependent on Pk, if a sequence of Pj, Pi1, Pi2, .., Pim, Pk exists. b. Pj is locally dependent on Pk, if above condition + Pj,Pk on same site. c. Each process maintains an array dependenti: dependenti(j) is true if Pi knows that Pj is dependent on it. (initially set to false for all i & j). Algorithm is as follows Sending the probe: if Pi is locally dependent on itself then deadlock. else for all Pj and Pk such that (a) Pi is locally dependent upon Pj, and (b) Pj is waiting on Pk, and (c) Pj and Pk are on different sites, send probe(i,j,k) to the home site of Pk. Receiving the probe: if (d) Pk is blocked, and (e) dependentk(i) is false, and (f) Pk has not replied to all requests of Pj, then begin dependentk(i) := true; if k = i then Pi is deadlocked else for all Pm and Pn such that (a’) Pk is locally dependent upon Pm, and (b’) Pm is waiting on Pn, and (c’) Pm and Pn are on different sites, send probe(i,m,n) to the home site of Pn. end.

×