Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In International Conference on Distributed Computing Systems, Yokohama, June 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....an object suspected to be a part of a garbage cycle. Augusteijn [5] presents an algorithm based on an incremental three color mark sweep algorithm applied to a distributed object system. Juul et al. 15] propose another distributed version of the incremental mark sweep algorithm. Ladin and Liskov [20] propose an algorithm that relies on a logically centralized global garbage detection service. Hughes [14] describes an appealing algorithm where mark bits are replaced by timestamps. The key idea is that a garbage object s timestamp remains constant whereas a nongarbage object timestamp increases ....

R. Ladin and B. Liskov. Garbage collection of a distributed heap. In International Conference on Distributed Computing Systems, Yokohama, June 1992.

....three color mark sweep algorithm applied to a distributed object system. Juul [15] proposes another distributed version of 13 the incremental mark sweep algorithm. The protocol to determine the termination of the mark phase is described as a two phase commit protocol. Ladin and Liskov [19] propose an algorithm that relies on a logically centralized global garbage detection service. The service may be physically replicated for higher availability. Each site maintains a list of objects that have been referenced from other sites. These are treated as roots for local garbage ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In International Conference on Distributed Computing Systems, Yokohama, June 1992.

....The globally oldest redo time is called minredo ; at any time, any IN entry which is timestamped with a time older than minredo is garbage and may be deleted. Minredo is calculated using the ring based terminationdetection protocol described in [58] The reference flagging scheme described in [32] uses Hughes algorithm adapted to cope with unreliable hardware, network, and messages, and also loosely synchronized clocks. Heap segmentation is at the node granularity; intra node pointers are direct, but inter node pointers are indirect in order to allow node local relocation of objects. The ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In 12th International Conference on Distributed Computing Systems, pages 708--15. IEEE Computer Society, June 1992.

....The globally oldest redo time is called minredo ; at any time, any IN entry which is timestamped with a time older than minredo is garbage and may be deleted. Minredo is calculated using the ring based termination detection protocol described in [58] The reference flagging scheme described in [32] uses Hughes algorithm adapted to cope with unreliable hardware, network, and messages, and also loosely synchronized clocks. Heap segmentation is at the node granularity; intra node pointers are direct, but inter node pointers are indirect in order to allow node local relocation of objects. The ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In 12th International Conference on Distributed Computing Systems, pages 708-15. IEEE Computer Society, June 1992.

....thought to be part of a garbage cycle: no collaboration with other processes is required. Keywords: distributed systems, garbage collection, termination detection 1 Introduction With the continued growth of distributed systems, designers are turning their attention to garbage collection [35, 30, 24, 22, 23, 7, 26, 27, 25, 14, 31, 15, 33, 20, 28, 18], prompted by the complexity of memory management and the desire for transparent object management. The goals of an ideal distributed garbage collector are that: safety: only garbage should be reclaimed. completeness: all garbage, including distributed cycles, at the start of a collection cycle ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In International Conference on Distributed Computing Systems, Yokohama, June 1992.

....when the marking phase terminates) Third, if a process i crashes, the entire tracing eventually halts as the global minimum redo s algorithm will be stucked at i s redo value. Thus, the algorithm is not resilient to failures. 2.3.2. 2 Logically Centralized Tracing The idea of this variant [73] is to compute the global accessibility of objects on a single highly available centralized service. All the information manipulated by this service (possibly replicated for fault tolerance) is stored in stable storage. 2 Each process performs asynchronous local collections and communicates with ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Int. Conf. on Distributed Computing Systems (ICDCS'92), pages 708--715, Yokohama (Japan), June 1992. IEEE.

....distributed garbage collector[MS91] that is efficient does not collect inter node cycles either. Some tracing algorithms like those described in [Hug85] and [JJ92] the current prototype) require some form of distributed termination detection which can impact scalability. The technique outlined in [LL92] collects inter node cyclic garbage efficiently but requires a bound on message delays, and loose synchronization of clocks. Lang et al. LQP92] describe an algorithm that groups nodes for purposes of garbage collection. Distributed cycles within such groups are detected. Node failures are handled ....

R. Ladin and B. Liskov. Garbage Collection of a Distributed Heap. In proceedings of 12th International Conference on DCS, 1992.

....These algorithms are efficient and tolerate all kinds of communication and hardware failures. It should be noted that Reference Counting and Reference Listing techniques have one common drawback. They do not allow collection of cycles of garbage. There are various solutions for collecting cycles [LL92, MJ95, UB97], but they all add a considerable amount of messaging to the implementation of their algorithms. In all these distributed GC algorithms, the number of messages used to 3 collect garbage depends tightly on the number of remote references and on the number of pointer operations. In distributed ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Int. Conf. on Distributed Computing Sys., pages 708--715, Yokohama (Japan), June 1992.

....of distributed garbage cycles. Based on the paths transmitted, the centralized service builds the graph of inter site references, and detects garbage (including dead cycles) with a standard tracing algorithm. The centralized service informs LGCs of accessibility of objects. In a later paper [Ladin 1992] Ladin and Liskov simplify and correct the deficiencies of the above proposal, adopting Hughes algorithm and loosely synchronized local clocks. Hughes algorithm eliminates inter space cycles of garbage, thereby eliminating the need for an accurate computation of the paths and for the central ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Int. Conf. on Distributed Computing Sys., Yokohama (Japan), June 1992.

....analysis. Thus, the centralised service is able to deduce that object t and z are globally reachable from RA . This solution is cheaper than the previous one in terms of local computation but increases significantly the communication between spaces and the centralised service. In a later paper, Ladin and Liskov [1992] simplify and correct the deficiencies of the above proposal by adopting Hughes algorithm (see Section 6.2) and loosely synchronised local clocks. Hughes algorithm eliminates inter space cycles of garbage, thereby eliminating the need for an accurate computation of the paths and for the central ....

.... Reference Counting [Bevan, 1987] ffl [Dickman, 1991] ffi ffl ffl ffl [Piquer, 1991] ffl ffl ffl ffl [Goldberg, 1989] ffl Reference Listing [Shapiro et al. 1990] ffl ffl ffl ffl ffl ffl ffl [Birrell et al. 1993] ffl ffl ffl ffi ffl ffl ffl Tracing [Ali, 1984] ffl [Hughes, 1985] ffl ffl [Ladin and Liskov, 1992] ffl ffl [Lang et al. 1992] ffi ffl ffi ffl [Juul and Jul, 1992] ffl ffl ffi : the characteristic is not intrinsically achieved by the distributed GC ffl : the characteristic is intrinsically achieved by the distributed GC. 7 Conclusion Reference counting associates with each public object a ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Principles of Distributed Computing, pages 708--715, 1992.

....of distributed garbage cycles. Based on the paths transmitted, the centralized service builds the graph of inter site references, and detects garbage (including dead cycles) with a standard tracing algorithm. The centralized service informs LGCs of accessibility of objects. In a later paper [5] Ladin and Liskov simplify and correct the deficiencies of the above proposal, adopting Hughes algorithm and loosely synchronized local clocks. Hughes algorithm eliminates inter space cycles of garbage, thereby eliminating the need for an accurate computation of the paths and for the central ....

Ladin, R., and Liskov, B. Garbage collection of a distributed heap. In Int. Conf. on Distributed Computing Sys. (Yokohama (Japan), June 1992).

....section will show some heuristics to determine when to start a collection operation. 5.2 Related Work Distributed garbage collection algorithms generally follow one of the two techniques used for collecting a single address space: mark and sweep or reference counting. Mark and sweep algorithms [33, 34, 45] first choose an area of distributed memory (although not blocking processing) determine the distributed root set, and organize inter node references. Then they determine the transitive closure of the references from the root set. Everything else is garbage. Reference counting techniques [15, 13, ....

....However, they do not locate distributed cycles. Both kinds of algorithms generally have two phases. In the local phase, completely unreferenced objects are garbage collected. In the second phase inter node information is propagated through the network, identifying more garbage. Ladin and Liskov[33] gives an algorithm for collecting cyclical garbage using a client server model. Distributed references are tracked by the server, which can use them to determine where garbage is. For scaling to larger networks, the article proposes dividing the network into areas, each with its own server and a ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In 12th Int. Conf. on Distributed Computing Systems, pages 708--715. IEEE, 1992.

....Many solutions for distributed garbage collection have been suggested, but they are not suitable for large scale distributed systems because their target systems are small in scale or they cannot collect cyclic garbage. Especially, it is a challenging problem to collect distributed cyclic garbage. [Ladin and Liskov, 1992] uses a logically centralized service which kept all information about inter systems refer1 ences, and [Jull and Jul, 1992] uses a global comprehensive tracing algorithm. Both of them do not scale well because they require global synchronization. Neuman, 1992] uses a local lower bound instead of ....

....as garbage. These algorithms do not need global synchronization, but the overhead of finding a global lower bound is extremely heavy. Hughes, 1985] uses a distributed termination detection algorithm to find the global lower bound, but the algorithm was very expensive and did not scale well. [Ladin and Liskov, 1992] uses a logically centralized service which kept all information about inter systems references, but this algorithm did not scale well either. Neuman, 1992] uses timeouts and last referenceable timestapms for garbage collection. Objects whose last referenceable timestamp is not more recent than a ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. International Conference on Distributed Computing Systems, pages 708-715, 1992.

.... on multiprocessors [Appel et al. 1988] but on large or widely distributed systems, the number of possible locations for a reference becomes very large, substantially affecting the methods and expense of garbage collection [Liskov and Ladin, 1986; Eckart and LeBlanc, 1987; Schelvis, 1989; Ladin and Liskov, 1992] On the other hand, under the strict container model, objects may not be referenced when the variable containing them no longer exists objects have a lifetime corresponding exactly to that of the variable containing them. This enables compilers to explicitly deallocate all objects, which ....

Rivka Ladin and Barbara Liskov, "Garbage Collection of a Distributed Heap," In Proceedings of the Twelfth ICDCS, Yokohama, Japan, June 1992.

....(but possibly replicated) server of any pointers into and out of the node. The local collector is responsible for determining the connectivity between the incoming and outgoing references. Rudalics [Rudalics90] points out an error in the original algorithm that is corrected by Ladin and Liskov [LL92] using an adaptation of Hughes s time stamp algorithm. Their solution also uses the centralised server clock to simplify Hughes s termination algorithm. Lang, Queninnec, and Piquer [LQP92] propose a technique where spaces (or nodes) are grouped. Any unreachable objects completely within a group ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Proceedings of the International Conference on Distributed Computing Systems, IEEE Press, 1992.

....the information to detect inter site garbage cycles [BE86] However, the fixed site becomes a performance and fault tolerance bottleneck. Ladin and Liskov proposed a logically central but physically replicated service that tracks intersite references and uses Hughes s algorithm to collect cycles [LL92] The central service avoids the need for a distributed algorithm to compute the global threshold. However, cycle collection still depends on timely correspondence between the service and all sites in the system. Subgraph Tracing The drawbacks of global tracing can be alleviated by first ....

R. Ladin and B. Liskov. Garbage collection of a distributed heap. In Proc. ICDCS. IEEE Press, 1992.

....IDL compiler. Arbitrary application languages can be supported at the cost of retargeting the back end for that particular language. Distributed cycles of garbage pose the most important problem, and are one of our future directions of work. All known algorithms that can collect cycles [6, 7, 8, 9] are either not fault tolerant, not scalable, or both. On the other hand, incomplete (more conservative) techniques are usually more resilient to message failures but these benefits rely on the assumption that distributed cycles are relatively rare. 3 Distributed Shared Memory Distributed ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Int. Conf. on Distributed Computing Sys., pages 708--715, Yokohama (Japan), June 1992.

....of distributed garbage cycles. Based on the paths transmitted, the centralised service builds the graph of inter site references, and detects garbage (includin dead cycles) with a standard tracing algorithm. The centralised service informs LGCs of accessibility of objects. In a later paper [12] Ladin and Liskov simplify, and correct the deficiencies of, the above proposal, adopting Hughes algorithm and loosely synchronised local clocks. Hughes algorithm eliminates inter space cycles of garbage, thereby eliminating the need for for an accu rate computation of the paths and for the ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Int. Conf. on Distributed Computing Sys., Yokohama (Japan), June 1992.

....by colouring the object graph in situ [10, 9] or indirectly from the logs maintained by the log keeping mechanism. The contents of these logs may be used to reconstruct consistent representations of the overall object graph that can be traced locally, either by a conceptually centralised service [11], or by each site that is participating in GGD [4] Once it has been determined that all live objects have been accounted for, or that enough information has been collected to reconstruct a consistent representation of the object graph, garbage objects can then be safely identified and their ....

....of the object graph, garbage objects can then be safely identified and their resources reclaimed. This termination detection is often performed as a distinct phase [10, 9, 4] although using a conceptually centralised log keeping service obviates an explicit termination detection phase [11]. Moreover Tel has also shown how these two phases can be superimposed [18, pp.193 226] To increase concurrency, multiple GGD iterations may overlap and proceed concurrently, e.g. an approach using time stamps [9] makes it possible to interleave any (bounded) number of iterations. However, all ....

R. Ladin and B. Liskov. Garbage collection of a distributed heap. In ICDCS'92, pp. 708--715, Jun. 92.

....[Bis77, This research was supported in part by the Advanced Research Projects Agency of the Department of Defense, monitored by the Office of Naval Research under contract N00014 91 J 4136. HM92, YNY94, AGF95, MMH96, CKWZ96] This is also the approach taken in many distributed system [LQP92, LL92, ML94, FS96] Generational collectors are a variant of partitioned collection that use the ages of objects to optimize the collection of younger, smaller partitions [LH83] however, the age based heuristics are not applicable to persistent stores [Bak93] A major problem with partitioned ....

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In Proc. International Conference on Distributed Computing Systems. IEEE Press, 1992.

....Counting (2.3) 2.1) Bis77] LH83] Ali84,Juu90] Ves87] Wen79,Bev87] Bis77,SGP90] Piq91,SDP92] Circular Garbage Collection (for use with reference tracking) Object Migration Tracing with Timestamps Centralized Server Tracing in Groups Complementary Tracing (2. 5) Bis77,SGP90] Ves87] Hug85] [LL92] [LPQ92] Trial Deletion [Ali84,Juu90] Figure 2 1: Techniques discussed in this chapter rather involved algorithms into their key ingredients shows how they can be combined with other algorithms to obtain a better mix. Towards the end of the discussion of each technique, we point out some of its ....

....garbage in question. Another drawback is that the execution of the distributed algorithm to compute global minimum redo is costly and slows down the collection of garbage. 2.5. 5 Centralized Server This scheme makes use of a logically centralized service that tracks all inter node references [LL92]. The implementation of the service may actually be distributed, and may use replication for high availability and reliability, but it appears as if it were run by one server [LLSG90] Nodes communicate with the service to provide it with information of their outlists and translists (Section ....

[Article contains additional citation context not shown here]

R. Ladin, and B. Liskov. Garbage Collection of a Distributed Heap. Int. Conference on Distributed Computing Systems, pages 708--715, Yokohoma, Japan, Jun 1992.

....information to detect inter site garbage cycles [BE86] However, the fixed site becomes a performance and fault tolerance bottleneck. Ladin and Liskov proposed a logically central but physically replicated service that tracks inter site references and uses Hughes s algorithm to collect cycles [LL92] The central service avoids the need for a distributed algorithm to compute the global threshold. However, cycle collection still depends on timely correspondence between the service and all sites in the system. Subgraph Tracing The drawbacks of global tracing can be alleviated by first ....

R. Ladin and B. Liskov. Garbage collection of a distributed heap. In Proc. ICDCS. IEEE Press, 1992.

....ONG93] do not scale to very large heaps because the nonlocal nature of tracing causes random disk accesses. Therefore, large systems partition the heap into independently collectible areas [Bis77, YNY94, AGF95, MMH96, CKWZ96] This is also the approach taken in many distributed systems [LQP92, LL92, ML94, FS96] Generational collectors are a variant of partitioned collection that use the ages of ob This paper appeared in ACM SIGMOD International Conference on Management of Data, 1997. This research was supported in part by the Advanced Research Projects Agency of the Department of ....

R. Ladin and B. Liskov. Garbage collection of a distributed heap. In Proc. ICDCS. IEEE Press, 1992.

No context found.

Rivka Ladin and Barbara Liskov. Garbage collection of a distributed heap. In International Conference on Distributed Computing Systems, Yokohama, June 1992.

No context found.

Rivka Ladin, Barbara Liskov: "Garbage Collection of Distributed Heap". Proc. of the Twelveth Int. Conf. on Dist. Comp. Syst., Yokohama (Japan), Sep. 92.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC