Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology Laboratory for Computer Science, May 1977. Technical report MIT/LCS/TR-178.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of physical pages may be required to store a given set of objects. To allow for more flexible segment sizes, we use an 11 bit floating point representation for segment size which was originally proposed by fellow Aries researcher Jeremy Brown [Brown93 and is similar to the format used in ORSLA [Bishop77]. Each segment is divided into blocks of size 2 bytes where 0 # B # 63, so six bits are required to specify the block size. The remaining 5 bits specify the length 1 # L # 32 of the segment in blocks: the segment size is L2 . Note that the values 1 # L # 16 are only required when B = ....

Peter B. Bishop, "Computer Systems with a Very Large Address Space and Garbage Collection ", Ph.D. Thesis, Dept. of EECS, M.I.T., May 1977.

....adjusted to provide better responsiveness, though we have not yet explored the configuration space fully, or related it to characteristics of various benchmarks, to offer a tuning strategy. 5. Related Work The Beltway framework combines and exploits key insights of incremental garbage collection [6], segregating objects to different physical regions of the heap in order to improve collector (and sometimes mutator) performance. This section compares Beltway to other collectors with respect to object segregation, pointer tracking, promotion policies, incrementality, completeness, and hybrids. ....

Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, MIT Laboratory for Computer Science, May 1977. Technical report MIT/LCS/TR--178.

....can trigger garbage collections. Neither reference counting nor reference listing can reclaim cyclic garbage. Typically, systems relying on these mechanisms assume that cyclic garbage is infrequent and that any accumulation thereof can be tolerated. Hybrid technologies, such as object migration [13], address this shortcoming. Objects involved in cycles are migrated to other spaces until the cycle is locally contained in one space and can be collected. Ka#eOS s merging of user heaps into the kernel heap can be viewed as migration of a user heap s objects to the kernel heap. 6.2.2 Scheduling ....

Bishop, P. B. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology Laboratory for Computer Science, Cambridge, MA, May 1977.

....take advantage of the scalable nature of reference counting and try to augment it so that cyclic garbage can also be collected. Some techniques propose running a distributed tracing algorithm at a low frequency in addition to the reference counting techniques to reclaim cyclic garbage. Bishop [9] proposes an object migration technique where objects suspected of belonging to a garbage cycle are migrated to a single node where they can be collected by a cyclic local garbage collector. Rodriguez Rivera [32] proposes an algorithm based on reference listing augmented with back tracing. For an ....

Peter B. Bishop. Computer systems with a very large address space and garbage collection. MIT Report LCS/TR--178, Laboratory for Computer Science, MIT, Cambridge, MA., May 1977.

....schemes have an entry for every instance of a pointer in a given area; others have one entry for an area regardless how many copies of the pointer there are in that area. 22 ORSLA The first area based garbage collection system in the literature is the reference listing scheme described in [7] as part of the custom hardware capability system ORSLA. ORSLA has an IN and OUT entry (actually the same object, an Inter Area Link (IAL) threaded onto one area s IN list and the other area s OUT list) for every single inter area pointer. Whenever an inter area pointer is created, ORSLA s ....

....does not record incoming pointers from that area. ORSLA eliminates garbage cycles by migrating objects which aren t reachable from area local roots into areas that reach them; in theory, this eventually causes a garbage cycle to collapse to a single area, at which point area local GC destroys it. [7] doesn t work out details insuring that migration terminates. For many architectures, the overhead of an IAL for every inter area pointer would be unacceptable due to the consumption of per node physical memory; additionally, having one IAL object serve as both an IN and an OUT entry requires that ....

[Article contains additional citation context not shown here]

Peter B. Bishop. Computer Systems With A Very Large Address Space And Garbage Collection. PhD thesis, Massachusetts Institute of Technology, May 1977.

....registers to keep the trace order of constant list structure similar between flips. If the average size of an object is much larger than the size of a pointer, the working set may also be reduced by storing the orwardJng addresses in a separate table instead of in the old objects in f:omspace [16]. In a virtual memory environment, the active address space will automatically expand and contract in response to changes in the number of accessible cells if ) fLIP re adjusts the size of fromspace to (l.m) cells in tospace] just before interchanging the semispaces; and 2) flil.pi occurs ....

....for the algorithm s correctness or real time properties. The average properties of the mlgortthm when run on a virtual memory machine need to be extensively investigated. Tire space reqmred by our algorithm maybe excessive for some applications, Perhaps a synthesis of Bishop s ercr concept [16] with our method COuld reduce the memory requirements of a list processing system while presetvir the bounded time properties of the elementary operations. A garbae collection algorithm can be viewed as a means for converting a Von Neumann style random access memory (with side effects [ t] ....

[Article contains additional citation context not shown here]

Bishop, P. B. "Computer Systems with a Very Large Address Space and Garbage Gollection". Ph.D. Thesis, M.I.T. Department .of Electrical Engineering and Computer Science. Also TR-178, MIT LCS, May 1977.

....listing schemes have an entry for every instance of a pointer in a given area; others have one entry for an area regardless how many copies of the pointer there are in that area. ORSLA The first area based garbage collection system in the literature is the referencelisting scheme described in [7] as part of the custom hardware capability system ORSLA. ORSLA has an IN and OUT entry (actually the same object, an Inter Area Link (IAL) threaded onto one area s IN list and the other area s OUT list) for every single inter area pointer. Whenever an inter area pointer is created, ORSLA s ....

....not record incoming pointers from that area. ORSLA eliminates garbage cycles by migrating objects which aren t reachable from area local roots into areas that reach them; in theory, this eventually causes a garbage cycle to collapse to a single area, at which point area local GC destroys it. [7] doesn t work out details insuring that migration terminates. For many architectures, the overhead of an IAL for every inter area pointer would be unacceptable due to the consumption of per node physical memory; additionally, having one IAL object serve as both an IN and an OUT entry requires that ....

[Article contains additional citation context not shown here]

Peter B. Bishop. Computer Systems With A Very Large Address Space And Garbage Collection. PhD thesis, Massachusetts Institute of Technology, May 1977.

....in the development a high performance fault tolerant system. 26 2.2.4 ORSLA Several object oriented system architectures have been proposed that have had significant impact on subsequent work. Perhaps the best known of these is ORSLA. ORSLA (Object References in a Single Large Address Space) Bishop 77] introduces areas to increase the locality of reference among related objects by grouping them together. Bishop uses areas to develop an efficient garbage collector for objects in a single, very large, paged, linear address space. His ideas influenced several other designs, including those of ....

....to reclaim objects that become inaccessible. A number of garbage collection schemes were considered for Distributed Smalltalk. We initially considered an incremental garbage collector that would integrate local and system garbage collection. Of the incremental collectors examined [Dijkstra 78, Bishop 77, Baker, Jr. 78, Lieberman 80, Luniewski 79, Ungar 84, Kung 77] the Kung Song algorithm, as modified by Chansler for StarOS, appeared to exhibit the best fit with the rest of 82 the design. The correctness argument of the Kung Song Chansler algorithm is readily adapted to the requirements of ....

Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, MIT, May 1977.

....some kind of reference counting, which fails to detect distributed cycles of garbage. To avoid important memory leaks in distributed storage systems, detection of distributed cycles of garbage has become an active area of research. Two kinds of algorithms address the problem: percycle algorithms [2, 10, 7, 12, 11, 13], i.e. algorithms triggered to verify that a particular suspected object belongs to a free cycle, and all at once algorithms [8, 9, 5] i.e. algorithms which detect all free cycles in a single mechanism, sometimes together with acyclic garbage. Permission to make digital or hard copies of all or ....

....in time, since the veri er is erroneously triggered a lot of times, and suspecting too few objects is expensive in memory, since reclamation of real garbage is delayed. When an object is suspected, the veri er is executed. Proposed mechanisms are: 1) migrating suspected objects to a single site [2, 10], where a tracing local garbage collector can collect the cycle; 2) back tracking [7, 11, 13] i.e.tracing the references to the object (backward references) recursively until either a root is reached (if the object is erroneously suspected) or all backward references are traced without nding a ....

Bishop, P. B. Computer systems with a very large address space and garbage collection. MIT Report LCS/TR{178, Laboratory for Computer Science, MIT, Cambridge, MA., May 1977.

....cache and log updates for crash recovery. Schemes that trace the entire heap (e.g. Bak78, KW93, ONG93] do not scale to very large heaps because the non local nature of tracing would cause random disk accesses. Therefore, some systems partition the heap into independently collectible areas [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, ....

....Work Partitioned collection has much in common with independent collection in distributed systems. Therefore, we relate our work to systems with large heaps as well as large distributed systems. Bishop proposed dividing a large address space into independently collectible partitions in 1977 [Bis77] He proposed collecting cyclic garbage by migrating objects to partitions that reference them. Migration is also used in some distributed systems because it is fault tolerant and decentralized [SGP90, ML95] The cost of migration is copying objects and patching up the references to the moved ....

Peter B. Bishop. Computer systems with a very large address space and garbage collection. Technical Report MIT/LCS/TR--178, MIT, 1977.

....results in the following two variants: i) hybrid collectors, CHAPTER 2. OVERVIEW OF GC ALGORITHMS 37 and (ii) generational collectors. In the first variant each partition is traced independently from the others, and cross partition references are handled with a reference counting algorithm [19, 90]. This variant is very interesting for large distributed and or persistent systems and is described in more detail in Section 2.3. The second variant, generational collectors, aims at reducing the GC pause time by decreasing the amount of memory that has to be collected. For this purpose, it ....

....cross partition cycles of garbage. There are three solutions for this problem: an additional cross partition tracing algorithm, object migration, and trial deletion. Cross partition tracing algorithms are addressed in the next section. Object Migration The basic idea of the migration technique [19, 111] is the following: migrating several objects to a single partition (in a process) in order to transform a distributed cycle into a local cycle that can be easily reclaimed with a tracing algorithm. The problem with this technique is twofold: i) some objects cannot be migrated due to systems ....

Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology Laboratory for Computer Science, May 1977. Technical report MIT/LCS/TR-178.

....For example, Kolodner [4] discusses this problem in the context of his Argus design, but does not offer a satisfactory solution. We are unaware of any other solution. How best to update pointers that cross between two heaps is a common problem for distributed and generational garbage collectors [2, 5, 11]. Unfortunately, none of this work is applicable to the transaction commit problem because transaction commit happens when chosen by the client, and the set of objects that must be made newly persistent is defined by reachability from the persistent root at that time. In contrast, generational ....

Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology Laboratory for Computer Science, May 1977. Technical report MIT/LCS/TR-178.

....of garbage either requires an adaptation of distributed termination detection protocols(might not scale well) or it requires migrating objects belonging to a cycle to other nodes so that cycles will end in a single node and will be reclaimed there. It is clear that moving objects physically as in [Bis77] or logically as in [Ves87] is undesirable. We propose a scheme that extends the scheme given in [SGP90] naturally and without migrating objects; the existing scheme is used except that extra information is passed around and one extra type of message is used. The paper proceeds as follows. First, ....

P. B. Bishop. Computer systems with a very large address space and garbage collection. Tech. Rep. MIT/LCS/TR-178, M.I.T. Laboratory of Computer Science, Cambridge, MA, May, 1977. 2 An object accessible from at least one remote space is called public, as opposed to private objects/citePS92. 5

....cannot be sent a message, etc. What are algorithms for garbage 101 collection that can make use of massive parallelism Certainly the allocation of objects will greatly influence (and should be influenced by) garbage collection. Are existing algorithms (e.g. marksweep for large memories [Bis77] or generational scavenging algorithms [Ung84] useful Can locality improving considerations be folded into these algorithms For example, can the reference pattern influence the copying phase of a copying garbage collection. Another aspect of managing very large numbers of objects on very ....

Peter Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, MIT LCS, Cambridge, MA, May 1977.

....garbage collection (DGC) DGC is particularly important in a large scale distributed system, since a de allocation error might cause a completely unrelated program to fail, possibly far away and an arbitrarily long time later. DGC has been a subject of academic research since at least 1977 [4]. With the growth of the Internet, DGC is now receiving its share of commercial attention. For instance, both Java RMI [28] and DCOM [23] come with some form of DGC. The next section contains a quick review of DGC techniques; for a more indepth treatment, we refer to published surveys [1, 21] ....

P.B. Bishop. Computer systems with a very large address space and garbage collection. Technical Report MIT/LCS/TR-178, Mass. Insitute of Technology, Cambridge MA (USA), 1977.

....can be gained from the programming language system experience. One fundamental insight is that the performance of garbage collection over a large address space is improved if the objects in the address space are partitioned into groups, where storage in each group can be reclaimed independently [3], 22] Thus, only a subset of a potentially huge set of objects needs to be considered at any point by a collector. There are two key advantages to collecting a subset of the total object space: the locality of reference of the collection algorithm is greatly improved, substantially reducing ....

....[17] 26] Our work can be extended with these techniques. Partition based collection algorithms are an extension of generational algorithms used in primary memory garbage collection. In a 1977 dissertation, Bishop investigated the possibility of garbage collection in a very large address space [3]. Bishop suggested dividing a large address space into areas and maintaining lists of interarea references to allow each area to be collected independently. Lieberman and Hewitt applied Bishop s approach to conventional language systems such as Lisp [22] They enhanced Bishop s idea with the ....

P.B. Bishop, "Computer Systems with a Very Large Address Space and Garbage Collection," PhD thesis, Mass. Inst. of Technology Lab for Computer Science, Cambridge, Mass., May 1977.

....et al. 1992] Card tables as described tend to retain scattered marked entries over time, and it is better to summarize them at collection time (into a concise remembered set representation) so that the card table reflects only incremental updates [Hosking and Hudson, 1993] 2.1. 5 Mature spaces Bishop [Bishop, 1977] and later Hudson and Moss [Hudson and Moss, 1992] proposed dividing the heap into spaces managed by different algorithms. Bishop allows several independent areas, not necessarily all garbage collected, and uses lists of incoming and outgoing inter area links. Hudson and Moss divide the heap by ....

Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA, May 1977.

....discard remote references but it is also able to reclaim garbage as the local garbage collector. However, both garbage collectors are not entirely independent because of the potential overlap in their activity. 5. 2 Object Migration The key idea of the object migration technique (first proposed by Bishop [1977], is to move all objects of a garbage cycle into a single space, provided that a local tracing collector reclaims intra space cycles. Figure 10 illustrates the consolidation of a garbage cycle composed. Initially (i) objects x, y and z are distributed respectively on spaces A, B and C. Step (ii) ....

Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology Laboratory for Computer Science, May 1977. Technical report MIT/LCS/TR-178.

No context found.

Peter B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology Laboratory for Computer Science, May 1977. Technical report MIT/LCS/TR-178.

No context found.

P. B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, MIT Laboratory for Computer Science, May 1977.

No context found.

P. B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, Massachusetts Institute of Technology Laboratory for Computer Science, May 1977. Technical report MIT/LCS/TR-178.

No context found.

P. B. Bishop. Computer Systems with a Very Large Address Space and Garbage Collection. PhD thesis, MIT Laboratory for Computer Science, May 1977.

No context found.

Bishop, P. Computer systems with a very large address space and garbage collection. Tech. Rept. TR-l78 MIT Lab. for Computer Science, Cambridge, Mass., May 1977.

No context found.

Peter B. Bishop, "Computer Systems with a Very Large Address Space and Garbage Collection", Ph.D. Thesis, Dept. of EECS, M.I.T., May 1977.

No context found.

Bishop, P.B. Computer Systems with a very large address space and garbage collection. Ph.D. Thesis, TR-178, MIT Lab. for Comp. Sci., Camb., MA, May 1977.

First 50 documents  Next 50

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