K. A. M. Ali. Garbage Collection Schemes for Distributed Storage Systems. Proceedings of Workshop on Implementation of Functional Languages, pages 422--428, Aspenas, Sweden, February 1985.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....large storage does not allow timely collection of garbage. Most scalable systems therefore use some variant of distributed reference counting [Bis77] The variants differ in the information kept for incoming remote references. Some schemes only record a flag for remotely referenced local objects [Ali84, JJ92]. Although this approach minimizes the reference information, it cannot detect locally when an object ceases to be referenced remotely. Other schemes record a count of how many external nodes have references to an object [Ves87] These schemes can detect when an object is no longer remotely ....

K. A. M. Ali. Garbage Collection Schemes for Distributed Storage Systems. Proceedings of Workshop on Implementation of Functional Languages, pages 422--428, Aspenas, Sweden, February 1985.

....marking phase, only the global roots are marked. Each partition trace propagates marks from the root set of the partition to the outlist. The marking phase terminates when marks are known to have propagated fully through all partitions. Similar schemes have been used in some distributed systems [Ali85, Hug85] although these systems rely on global marking to collect both acyclic and cyclic interpartition garbage. Other schemes either propagate global marks separately from regular partition traces [JJ92] or are not guaranteed to terminate correctly in the presence of concurrent mutations ....

K. A. M. Ali. Garbage collection schemes for distributed storage systems. In Proc. of Workshop on Implementation of Functional Languages, pages 422--428, 1985.

....Tracking Reference Flagging Reference Counting Weighted Reference Counting Reference Listing Indirection, and Strong Weak Ptrs (2.2) 2.4) HK82] GC in Separate Areas Inter Area Reference Counting Generational Collection Single Site Tracing and Reference 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] ....

....[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 drawbacks. In doing this, different ....

[Article contains additional citation context not shown here]

K. A. M. Ali. Garbage Collection Schemes for Distributed Storage Systems. Proceedings of Workshop on Implementation of Functional Languages, pages 422--428, Aspenas, Sweden, Feb 1985.

....garbage independently of other sites. However, for a local trace to be safe, object references from other sites must be treated as roots. Thus, many distributed systems use local tracing in combination with some variant of inter site reference counting to track inter site references [Bis77, Ali85, Bev87, SDP92, JJ92, BEN 93, ML94] Local tracing has the desirable locality property that col This research was supported in part by the Advanced Research Projects Agency of the Department of Defense, monitored by the Office of Naval Research, contract N00014 91 J 4136. This paper appears ....

....in collecting an inter site garbage cycle is to preserve locality, that is, to involve only the sites containing the cycle. This has proven surprisingly difficult. Most previous schemes do not preserve locality. For example, some conduct complementary global traces in addition to local tracing [Ali85, JJ92] The drawbacks of global tracing can be alleviated by tracing within groups of selected sites [LQP92, MKI 95, RJ96] but inter group cycles may never be collected. Few schemes for collecting inter site cycles have the locality property. The most prominent among these is based on ....

[Article contains additional citation context not shown here]

K. A. M. Ali. Garbage collection schemes for distributed storage systems. In Proc. Workshop on Implementation of Functional Languages, pages 422--428, 1985.

....can collect any garbage. Such a trace would not be timely or fault tolerant because every site must wait until the global trace has finished. Timeliness and fault tolerance require that a site trace local objects independently of other sites. This is the approach taken in many distributed systems [Ali85, Bev87, SDP92, JJ92, BEN 93, ML94] Local tracing minimizes inter site dependence. In particular, garbage that is not referenced from other sites is collected locally. Further, a chain of garbage objects spanning multiple sites is collected through cooperation within the sites holding the ....

....Thus, global marking might never complete in a large system. The challenge in collecting an inter site garbage cycle is to preserve the locality property, that is, to involve only the sites containing the cycle. This has proven difficult: most previous schemes do not preserve locality [Ali85, JJ92, LQP92, MKI 95, RJ96] The few that do seem prohibitively complex or costly [SGP90, Sch89, LC97] We present the first practical scheme with locality to collect inter site cycles. It has two parts. The first part identifies objects that are highly likely to be cyclic garbage the ....

[Article contains additional citation context not shown here]

K. A. M. Ali. Garbage collection schemes for distributed storage systems. In Proc. Workshop on Implementation of Functional Languages, pages 422--428, 1985.

....reference counting is preferred for systems with large numbers of nodes because it is more faulttolerant and scalable, and quicker at collecting distributed garbage. Many variants of distributed reference counting schemes have been proposed to enhance fault tolerance and reduce overheads [Ali84, Bev87, Ves87, Piq91, LL92, SDP92, BENOW93, ML94]. Distributed reference counting algorithms cannot collect multi node cycles of garbage objects. This is particularly undesirable in long lived systems such as persistent object stores, where even small amounts of uncollected garbage can accumulate over time to cause a significant storage loss. ....

....of the Principles of Distributed Computing, 1995. 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. using a complementary marking scheme to collect cyclic garbage [Ali84, JJ92, LQP92], or by migrating objects so that cyclic garbage ends up in a single node and is collected by the local collector [Bis77, SGP90, GF93] The advantage of migration is that, like distributed reference counting, it is decentralized and fault tolerant. The collection of a cycle requires the ....

[Article contains additional citation context not shown here]

K. A. M. Ali. Garbage Collection Schemes for Distributed Storage Systems. Proceedings of Workshop on Implementation of Functional Languages, pages 422--428, Aspenas, Sweden, February 1985.

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