P. Ferreira and M. Shapiro, "Modelling a Distributed Cached Store for Garbage Collection: The Algorithm and Its Correctness Proof," Proc. Eighth European Conf. Object-Oriented Programming, July 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

No context found.

P. Ferreira and M. Shapiro, "Modelling a Distributed Cached Store for Garbage Collection: The Algorithm and Its Correctness Proof," Proc. Eighth European Conf. Object-Oriented Programming, July 1998.

No context found.

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection: the algorithm and its correctness proof. In ECOOP'98, Proc. of the Eight European Conf. on Object-Oriented Programming, Brussels (Belgium), July 1998.

No context found.

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection: the algorithm and its correctness proof. In ECOOP'98, Proc. of the Eight European Conf. on Object-Oriented Programming, Brussels (Belgium), July 1998.

....1) this may lead to the reclamation of the master replica; however, a replica still exists. This situation arises because the Java distributed collector is not aware of replicas; for the collector, these are simply di erent objects. Once again, we plan to extend our previous work in this area [5, 6]. ....

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection: the algorithm and its correctness proof. In ECOOP'98, Proc. of the Eight European Conf. on Object-Oriented Programming, Brussels (Belgium), July 1998.

....We use this functionality to provide a service of WebNews where dangling references and storage leaks never occur. In addition, webmasters are released from the complex task of memory management. The system uses a distributed garbage collection algorithm to maintain the referential integrity [ferreira98]. The paper is organised as follows. In the next section we present the model in which our approach is based. Then we describe the system architecture. The fourth section presents the most important implementation aspects. Section five presents our conclusions and some future work. 2. Model ....

....of references; this results either in the creation of new references or in the deletion of already existing ones. Figure 2: Example of new content creation based on imported content 3. 3 Garbage Collection Referential integrity is enforced through a distributed garbage collection algorithm [ferreira98]. We do not describe the algorithm in detail since it is out of the scope of this paper. Instead, we only present its main ideas. As mentioned in the previous section, in the worldwide news gathering model, assignments are the operations that create references to information objects. An ....

[Article contains additional citation context not shown here]

Ferreira, Paulo and Shapiro, Marc, ,1998. Modelling a Distributed Cached Store for Garbage Collection: the algorithm and its correctness proof, ECOOP'98, Brussels, Belgium, July 1998

....of collection in a DSM, which Larchant satis es. The algorithm is safe (no reachable data is reclaimed) and live (garbage is eventually reclaimed) Sadly, but unavoidably, it is not complete. The underlying theoretical model and a proof of correctness can be found in Ferreira and Shapiro [8]. More information about the Larchant algorithm can be found in previous publications [6, 7, 25] 4.2 GC algorithm and safety rules The Larchant memory, as shown in Figure 3, is subdivided into coarse grain bunches (typically, a set of continguous pages) in which objects are allocated. A bunch ....

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In 12th Euro. Conf. on Object-Oriented Prog. (ECOOP), Brussels (Belgium) , July 1998. http://www-sor.inria.fr/publi/MDCSGC ecoop98.html.

....be generalised to arbitrary objects. 4.2 Distributed Garbage Collection (DGC) Persistence in PerDiS is based on reachability [3] To decide whether an object is reachable is a dicult problem in a distributed and replicated environment. We use the Larchant distributed GC model and algorithm [9, 14]. A DGC is modeled as a per site constructive component, which identi es new pointers, a per site destructive component, which identi es unreachable objects, and a distribution component, which ensures consistency of GC meta data. Larchant posits ve ordering rules that enforce a safe global ....

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In 12th Euro. Conf. on Object-Oriented Prog. (ECOOP), Brussels (Belgium) , July 1998. http://www-sor.inria.fr/publi/MDCSGC ecoop98.html.

....pointers (caused by an application program erroneously deleting a reachable object) would make the store unsafe, causing programs to fail unexpectedly, possibly years after the error. In PerDiS instead, storage is managed automatically, using the Larchant distributed garbage collection algorithm [19,20]. Larchant is based on meta information that supplements pointers, called stubs and scions. A stub describes a pointer that points out of a cluster, and a scion a pointer into a cluster. Stubs and scions are also used by the swizzler. The algorithm is divided into a constructive and a ....

....garbage collector must be aware of caching and concurrency, because an object might not be reachable at a particular site but still be reachable globally. It must also carefully order its actions because of possible global race conditions. This is explained in more detail in the Larchant articles [19,20]. 4.5 Security The PerDiS platform protects whole clusters according to user speci ed access rights and provides secure communication between PDs. Communication is secured against eavesdropping, impersonation, tampering and replay by attackers from within or outside the VE. The VE is composed of ....

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In Proc. of the 12th European Conf. on Object-Oriented Programming (ECOOP), Brussels (Belgium), July 1998.

....it can access. This property, called referential integrity, is essential for program correctness; it also improves performance by removing any need for run time checks. Furthermore, GC avoids memory leaks by automatically deallocating unused memory. Our system uses the Larchant GC algorithm [8, 9], designed for a shared, replicated memory. It is distributed and minimizes induced I O, communication and synchronization costs. Larchant uses partitioned GC [5] also called hybrid GC [17] with reference listing for inter partition references, and tracing for intra partition references. A ....

....to ask the home site, which is fixed and well known. The home site does maintain the probOwner information and is able to forward requests correctly. This is because Larchant applies the so called Union Rule, whereby an object is garbage only if it is unreachable in the union of all replicas [9]. 3.1 Reference listing and tracing Larchant runs a tracing GC algorithm on every site. The trace can use either mark and sweep or copy collection [12, 25] It collects each cluster independently from others; furthermore, each replica of a cluster is traced independently of other replicas. The ....

[Article contains additional citation context not shown here]

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In European Conference on Object-Oriented Programming (ECOOP), Brussels (Belgium), July 1998. http://www-sor.inria.fr/publi/MDCSGC ecoop98.html.

....DGC algorithm satisfies five simple correctness rules. The algorithm is safe (no reachable data is reclaimed) and live (garbage is eventually reclaimed) Sadly but unavoidably, it is not complete. The underlying theoretical model and a proof of correctness can be found in Ferreira and Shapiro [7]. More information about the Larchant algorithm can be found in previous publications [5, 6, 24] 4.2 GC algorithm and safety rules The Larchant memory, as shown in Figure 3, is subdivided into coarse grain bunches (typically, a set of continguous pages) in which objects are allocated. A bunch ....

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In 12th Euro. Conf. on Object-Oriented Prog. (ECOOP), Brussels (Belgium), July 1998. http://www-sor.inria.fr/publi/MDCSGC ecoop98.html.

....pointers (caused by an application program erroneously deleting a reachable object) would make the store unsafe, causing programs to fail unexpectedly, possibly years after the error. In PerDiS instead, storage is managed automatically, using the Larchant distributed garbage collection algorithm [17, 18]. Larchant is based on meta information that supplements pointers, called stubs and scions. A stub describes a pointer that points out of a cluster, and a scion a pointer into a cluster. Stubs and scions are also used by the swizzler. The algorithm is divided into a iconstructivej and a ....

....garbage collector must be aware of caching and concurrency, because an object might not be reachable at a particular site but still be reachable globally. It must also carefully order its actions because of possible global race conditions. This is explained in more detail in the Larchant articles [17, 18]. 4.5 Security The PerDiS platform protects whole clusters according to user speci ed access rights and provides secure communication between PDs. Communication is secured against eavesdropping, impersonation, tampering and replay by attackers from within or outside the VE. The VE is composed of ....

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In 12th Euro. Conf. on Object-Oriented Prog. (ECOOP), Brussels (Belgium), July 1998. http: //www-sor.inria.fr/publi/MDCSGCecoop98.html.

....it can access. This property, called referential integrity, is essential for program correctness; it also improves performance by removing any need for run time checks. Furthermore, GC avoids memory leaks by automatically deallocating unused memory. Our system uses the Larchant GC algorithm [8, 9], designed for a shared, replicated memory. It is distributed and minimizes induced I O, communication and synchronization costs. Larchant uses partitioned GC [5] also called hybrid GC [17] with reference listing for inter partition references, and tracing for intra partition references. A ....

....and Shapiro [7] A given cluster can be replicated at any number of sites. Larchant remains correct even if the replicas are not mutually coherent. This is because Larchant applies the so called Union Rule, whereby an object is garbage only if it is unreachable in the union of all replicas [9]. 3.1 Reference listing and tracing Larchant runs a tracing GC algorithm on every site. The trace can use either mark and sweep or copy collection [12, 25] It collects each cluster independently from others; furthermore, each replica of a cluster is traced independently of other replicas. The ....

[Article contains additional citation context not shown here]

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In European Conference on Object-Oriented Programming (ECOOP), Brussels (Belgium), July 1998. http://www-sor.inria.fr/publi/MDCSGC ecoop98.html.

No context found.

Paulo Ferreira and Marc Shapiro. Modelling a distributed cached store for garbage collection. In Jul [Jul1998].

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