Umesh Maheshwari. Distributed garbage collection in a client--server, transactional, persistent object system. Technical Report MIT/LCS/TR--574, MIT Press, February 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... GC Although I shall discuss most of the major approaches to the distributed GC problem below, additional approaches and references are presented in the survey paper [56] Another presentation and detailed analysis of several distributed GC schemes is presented in chapter 2 of the thesis [43]. Additional references may be found in the bibliography paper [61] which includes not only avant garde topics such as parallel and distributed GC, but also a great many references on classical GC techniques. 21 2.2.2 Copying GC System wide copying GC doesn t actually use areas to any great ....

....collection has been the subject of much research. Thor runs on a heterogeneous collection of conventional hardware; object references are opaque (indirect) thus allowing intra area object relocation without requiring inter area communications. 23 The garbage collection scheme described in [43] uses conservative reference lists between operating regions (ORs) of the database itself, and between ORs and front end (FE) clients. The scheme is designed to be fault tolerant in the face of lost or delayed messages. Each OR maintains a table of objects reachable from other ORs, listing for ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Master of engineering thesis, Massachusetts Institute of Technology, February 1993.

.... GC Although I shall discuss most of the major approaches to the distributed GC problem below, additional approaches and references are presented in the survey paper [56] Another presentation and detailed analysis of several distributed GC schemes is presented in chapter 2 of the thesis [43]. Additional references may be found in the bibliography paper [61] which includes not only avant garde topics such as parallel and distributed GC, but also a great many references on classical GC techniques. 2.2.2 Copying GC System wide copying GC doesn t actually use areas to any great ....

....garbage collection has been the subject of much research. Thor runs on a heterogeneous collection of conventional hardware; object references are opaque (indirect) thus allowing intra area object relocation without requiring inter area communications. The garbage collection scheme described in [43] uses conservative reference lists between operating regions (ORs) of the database itself, and between ORs and front end (FE) clients. The scheme is designed to be fault tolerant in the face of lost or delayed messages. Each OR maintains a table of objects reachable from other ORs, listing for ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Master of engineering thesis, Massachusetts Institute of Technology, February 1993.

....An object that has moved leaves behind a surrogate containing its new xref. An example is shown in Figure 6.2. Server surrogates are eventually removed during server garbage collection after all objects that referred to the moved object by its old xref have been modified to contain its new xref[46]. Since objects can move from server to server, a persistent object also has a system wide unique identifier (OID) assigned to it. The servers store their persistent objects in very large disk segments. Tentatively, we have chosen a segment size of 64 kilobytes (64K) though this is a subject of ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object store. Technical Report MIT/LCS/TR-574, Massachusetts Institute of Technology, Laboratory for Computer Science, October 1993. Master's thesis.

.... as prefetched objects instead of a client sending explicit messages and waiting for them (assuming that the prefetching strategy prefetches objects referenced at the same server by a fetched object) Fewer external references are beneficial for the distributed garbage collection algorithm also [Mah93]. Interserver references can be reduced by moving an object to a server where most references to it 18 reside or vice versa. Apart from reducing inter server references, clustering an application s objects at a single server has the advantage that a 2 phase commit protocol is not needed to commit ....

....they are likely to be accessible when needed. Thor provides a persistent root for the object universe. An object becomes persistent if it becomes accessible from the persistent root. If an object becomes unreachable from the root, its storage is reclaimed by a distributed garbage collector [Mah93]. Each object has a globally unique id called oid . Objects contain data and references to other Thor objects. They are encapsulated so that an application using Thor can access their state only by calling their methods. In this thesis, we assume that the set of methods available to users are just ....

Maheshwari U. Distributed Garbage Collection in a Client-Server, Transactional Persistent Object System. S.M. thesis, Massachusetts Institute of Technology, 1993.

....considers processes communicating by messages (without shared memory) using a hybrid of tracing and counting. Each process traces its internal pointers; references across process boundaries are reference counted as they are sent in messages. Some object oriented databases use a similar approach [1, 23, 16]. In order to collect cycles of garbage, Lang, Queinnec and Piquer [14] augment a hybrid algorithm with a scan of dynamically changing groups of processes. Their groups are remote and entail a complex joining disbanding protocol and complex synchronization inside a group. Our grouping algorithm is ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Technical Report MIT/LCS/TM-574, Mass. Inst. of Technology, Lab. for Comp. Sc., Cambridge, MA (USA), October 1993.

....before such a comparison can be made, it is necessary to understand both how to manage an object cache and what the important parameters are for its performance. Copying objects to a client cache and managing that cache introduce new problems. Others have addressed aspects of storage management [48] and concurrency control [1] The thesis addresses the following three performance related problems: ffl implementing inter object references; ffl bringing objects into the client cache; and ffl evicting objects from the client cache. The most important results are the following: 1. Varying the ....

....which of its objects are present in each client cache. This information allows the server to invalidate objects in a client cache that have been made obsolete by updates at the server[46] The same information is also used for efficient concurrency control [1] and distributed garbage collection [48]. The server s information always represents a superset of the objects actually in the client s cache: no object can be in the client s cache without being recorded at the server, but the server may record an object that is not in use at the client. Each fetch request from the client causes the ....

[Article contains additional citation context not shown here]

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Technical Report MIT/LCS/TR-574, MIT Laboratory for Computer Science, 1993.

....shared memory) using a hybrid of tracing and counting. Each process traces its internal pointers; references across process boundaries are reference counted as they are sent in messages. Some object oriented databases use a similar approach [1, 25] In particular, Thor is a research OODB with PBR [19]. In Thor, the data resides at a small number of servers and is cached at workstations for processing. A Thor server counts references contained in objects cached at a client; Thor delays scanning as proposed in Section 4.1. On top of a hybrid algorithm, Lang, Queinnec and Piquer [17] propose ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Technical Report MIT/LCS/TM-574, Mass. Inst. of Technology, Lab. for Comp. Sc., Cambridge, MA (USA), October 1993.

....pointers; references across process boundaries are reference counted as they are sent in messages. Some object oriented databases use a similar approach [1, 6, 18] i.e. a partition can be collected independently from the rest of the database. In particular, Thor is a research OODB with PBR [14]. In Thor, the data resides at a small number of servers and is cached at workstations for processing. A Thor server counts references contained in objects cached at a client; Thor delays the creation of scions as proposed in Section 3.3. On top of a hybrid algorithm, Lang, Queinnec and Piquer ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Technical Report MIT/LCS/TM574, Mass. Inst. of Technology, Lab. for Comp. Sc., Cambridge, MA (USA), October 1993.

....considers processes communicating by messages (without shared memory) using a hybrid of tracing and counting. Each process traces its internal pointers; references across process boundaries are reference counted as they are sent in messages. Some object oriented databases use a similar approach [1, 22, 15]. In order to collect cycles of garbage, Lang, Queinnec and Piquer [13] augment a hybrid algorithm with a scan of dynamically changing groups of processes. Their groups are remote and entail a complex joining disbanding protocol and complex synchronization inside a group. Our grouping algorithm is ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Technical Report MIT/LCS/TM-574, Mass. Inst. of Technology, Lab. for Comp. Sc., Cambridge, MA (USA), October 1993.

....for each participant to complete all work expected of it. If an insert message to an owner fails, the coordinator will not receive an acknowledgment from it, and the transaction will abort as desired. Commits that involve insert messages thus have an added latency of one additional message. [Mah93] describes a scheme that hides this latency by having the coordinator send the insert messages on behalf of the participants in parallel with the prepare messages. However, this technique has drawbacks of its own it complicates the inorder delivery of insert and outlist messages, and may send ....

U. Maheshwari. Distributed Garbage Collection in a Client-Server, Transactional, Persistent Object System. Technical Report MIT/LCS/TR-574, Massachusetts Institute of Technology, July 1993.

....This protocol uses a proxy server per client that must be contacted whenever the client approaches a new server. This approach is not suitable if clients browse over many different servers; a light weight protocol is desirable in that case. We have studied one such protocol based on leases [Mah93b] Here, a client must periodically renew leases with servers that have stored information for it. A ban protocol may be applied to fields outside of garbage collection where it is necessary for all servers to ban a client before discarding state about the client, e.g. in on line financial ....

U. Maheshwari. Distributed garbage collection in a client--server, transactional, persistent object system. Technical Report MIT/LCS/TR--574, MIT Press, 1993.

....the identities of the individual source sites [Bev87, Piq91] When the count of a reference reduces to zero, it is removed from the inset. On the other hand, our scheme effectively records the identities of the source sites for each reference in the inset, a technique known as reference listing [Mah93a, BEN 93] Reference listing keeps more information than reference counting, but we use it for the following reasons: 1. Reference listing tolerates permanent site failures. If a client or a server X fails permanently, other servers need simply remove their inlists from X . With reference ....

U. Maheshwari. Distributed garbage collection in a client--server persistent object system. In E. Moss, P. R. Wilson, and B. Zorn, editors, OOPSLA/ECOOP '93 Workshop on Garbage Collection in Object-Oriented Systems, 1993.

....client requests (since it cannot be sure that the OR still retains its OR session information) although it can delay closing the client session in the hopes that it will reestablish connection with the OR and discover that the OR has not closed the OR session. More information can be found in [24]. It allows client sessions to survive failovers at ORs and avoids the need to maintain session information on stable storage at ORs. 3.2.2 Buffer Management Unlike other systems, Thor manages its buffers in terms of objects. Rather than bringing in a page when an object is fetched, we bring in ....

Umesh Maheshwari. Distributed garbage collection in a client-server, transactional, persistent object system. Master's thesis, Massachusetts Institute of Technology, 1993.

....so references that cross OR boundaries are rare. Thor provides a persistent root for the object universe. An object becomes persistent when it becomes reachable from the persistent root. If an object becomes unreachable from the root, its storage is reclaimed by a distributed garbage collector[12]. Client programs run at client workstations that are typically distinct from the servers that run ORs. For each client program, there is a front end, or FE, running on the client s machine. The FE provides an interface for the client to access Thor. To ensure consistency despite failures and ....

Maheshwari, U. Distributed garbage collection in a client-server, transactional, persistent object system. Tech. Rep. MIT/LCS/TR-574, Massachusetts Institute of Technology, 1993.

No context found.

Umesh Maheshwari. Distributed garbage collection in a client--server, transactional, persistent object system. Technical Report MIT/LCS/TR--574, MIT Press, February 1993.

No context found.

Umesh Maheshwari. Distributed garbage collection in a client--server persistent object system. In Moss et al. [Moss et al.1993].

No context found.

Umesh Maheshwari. Distributed garbage collection in a client--server transaction system. Master's thesis, Department of Electrical Engineering and Computer Science, Massachuset ts Institute of Technology, 1992.

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