M. Schelvis and E. Bledoeg. The implementation of a distributed Smalltalk. Lecture Notes in Computer Science, 322:212--232, 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....in CST. CST is thus an object oriented concurrent language [22] HoME is just an objectoriented language in which processes can be executed in parallel. CST is implemented on a uniprocessor, not a multiprocessor. Examples of Smalltalk which can be used on a distributed environment are [3] 5] [16]. These systems allow objects on different machines to be connected by a network for sending and responding to messages. We did not design HoME on the premise that HoME can be executed on a distributed environment. 7 Concurrent Programming in HoME Section 2 discussed concurrent programming in ....

Marcel Schelvis and Eddy Bledoeg, The Implementation of a Distributed Smalltalk, Proc. of ECOOP'88, 212-232, 1988.

....Smalltalk. Whereas Decouchant and the system of Schelvis and Bledoeg (Oc e Netherland) extended the Smalltalk virtual machine, the other projects chose to add proxy and message objects at the virtual image level. synchronization. fault tolerance. Availability: References: 27] 77] 159] 168] [189] 2.40 Distributed Smalltalk Process Developer: Description: Extension of Goldberg and Robson s Smalltalk (see section 2.92) oo. See Smalltalk 80 (section 2.92) memory model. parallelism. See Smalltalk 80 (section 2.92) The fork message can handle the node number. scheduling. ....

Marcel Schelvis and Eddy Bledoeg. The implementation of a Distributed Smalltalk. In Proc. of the European Conf. on Object-Oriented Programming, ECOOP'88, number 322 in Lecture Notes in Computer Science, pages 212--232, Oslo, Norway, August 15--17, 1988. Springer-Verlag Berlin, Heidelberg, New York.

.... Harold Carr Gamma carr cs.utah.edu Distributed Eiffel [103] Phi Phi Delta Delta H H A A Phi Phi Delta Delta H H A A Phi Phi Delta Delta H H A A Phi Phi Delta Delta H H A A m y t i v i t c a y r a d n u o b Distributed Smalltalk Object [31, 84, 172, 181, 208] Phi Phi Delta Delta H H A A Phi Phi Delta Delta H H A A m y t i v i t c a y r a d n u o b 46 Distributed Smalltalk Process [157] Phi Phi Delta Delta H H A A Phi Phi Delta Delta H H A A m y t i v i t c a y r a d n u o b DoPVM ....

Marcel Schelvis and Eddy Bledoeg. The implementation of a Distributed Smalltalk. In Proc. of the European Conf. on Object-Oriented Programming, ECOOP'88, number 322 in Lecture Notes in Computer Science, pages 212--232, Oslo, Norway, August 15--17, 1988. SpringerVerlag Berlin, Heidelberg, New York.

....the latency must be limited by a known small constant; not only relative to the total cost. In distributed systems cross node references complicates garbage collection, especially when some nodes are temporaryly unreachable. However many references tend to be short lived and local [Lieberman 83, Schelvis 88, Jul 88b] so dividing garbage collection in a nodelocal and a global collector may help this. The global garbage collection must still handle all cross node references and retain garbage not referenced from any node, even if the nodes are not running simultaneously. The object oriented approach ....

Marcel Schelvis and Eddy Bledoeg. The implementation of Distributed Smalltalk. In S. Gjessing and K. Nygaard, editors, ECOOP'88, European Conference on Object-Oriented Programming, Proceedings published in: Lecture Notes in Computer Science 322, pages 212--232, Springer-Verlag, Oslo, Norway, August 1988.

....than an instruction cycle. This is particularly the case for RISC processors. For efficiency, avoiding the processor idling while waiting for a response is vital. In most evaluations of distributed protocols, communication overhead is the principal metric. Published measurements [Bennett, 1987; Schelvis and Bledoeg, 1988] indicate remote cell access to be slower than local by two orders of magnitude. According to Arvind and Iannucci [1987] two main issues must be addressed by a successful exploitation of multiple processors: latency and synchronization. In general, processes that arrange their activities so as ....

....with their own local collector. His performance metric was communication overhead. Bennett reported that on aver29 age remote messages are slower than local messages by a factor of 1000. The measured system consisted of two Sun 2 diskless workstation connected by a 10 Megabit Second Ethernet. Schelvis Bledoeg [1988] also evaluated collectors for distributed Smalltalk implemented on a network of SUN workstations running Berkeley UNIX. These included a generation scavenging conservative distributed collector. Schelvis Bledoeg conclude that remote send is approximately 500 times slower than local send. In ....

Schelvis M and Bledoeg E (1988) The implementation of a distributed Smalltalk, ECOOP Proceedings, August 1988, LNCS 322, 212-32.

....in the decentralized system. 3. A host local timestamp of sufficient resolution. Such an identifier would be unique and would also tell where the object was created, which could be used as a good hint of location. Smalltalk has been used as the implementation basis for decentralized systems [Decou86, Benne87, McCul87, Schel88]. Each location is then a single user workspace. Various forms of forwarding objects (proxys) are used for remote addressing. Except for the system described by McCullough [McCul87] objects generally do not maintain strong (surrogate based) identity between different locations. The reason being, ....

....decentralized system which also interacts with a transaction mechanism. It then makes sense to try to unify the two as far as possible. It also makes sense to find means to avoid using the expensive decentralized and persistent identification mechanism for all objects. 1 The system described in [Schel88] does provide global consistency for replicated objects using the so called Thomas write rule [Thoma79] However these objects are special and only a small part of the total object space. A.Bjornerstedt 5 Using location independent identifiers for persistent objects means that there must be at ....

[Article contains additional citation context not shown here]

M. Schelvis and E. Bledoeg, "The Implementation of a Distributed Smalltalk," ECOOP'88 European Conference on Object-Oriented Programming, pp. 212-232, Oslo, Norway, Aug. 1988. A.Bjornerstedt 33

....pages fast enough so that applications rarely have to wait to allocate a new page; and, also, that it collects all inaccessible pages eventually. DISTRIBUTED GARBAGE COLLECTION Most garbage collectors proposed for distributed systems require double indirection for remote references [12, 16, 17]. This mechanism requires direct mutator collaboration, and imposes a severe overhead on the mutators when creating references and, especially, when dereferencing. Some collectors are designed for systems in which pages do not migrate [12, 17] or in which page migration requires mutator ....

....require double indirection for remote references [12, 16, 17] This mechanism requires direct mutator collaboration, and imposes a severe overhead on the mutators when creating references and, especially, when dereferencing. Some collectors are designed for systems in which pages do not migrate [12, 17], or in which page migration requires mutator collaboration to update some of the collector s data structures [4] Many proposals require global synchronization among processors [2, 3, 4, 9, 13] which in some cases implies halting the mutators. Finally, in some parallel or distributed systems ....

[Article contains additional citation context not shown here]

Marcel Schelvis and Eddy Bledoeg. 1988 (August). The implementation of a distributed Smalltalk. In Gjessing, Stein and Kristen Nygaard, editors, ECOOP'88: European Conferece on ObjectOriented Programming (LNCS 322), pages 212--232, Berlin. Springer-Verlag.

....EZ system that is subject to frequent idle periods, but an off line approach is unsuitable for a distributed system. Distributed EZ will use a distributed garbage collector that works in concert with the shared virtual memory manager described above. It is a distributed mark and sweep collector [4, 6, 24], and it is concurrent and real time. Technically, algorithms based on reference counting [10, 17] are more efficient, but require additional data for every pointer that might refer to a page on another processor, or additional synchronization between subsets of the processors at each reference. ....

M. Schelvis and E. Bledoeg. The implementation of a distributed Smalltalk. In S. Gjessing and K. Nygaard, editors, ECOOP'88: European Conference on ObjectOriented Programming (LNCS 322), pages 212--232, Berlin, Aug. 1988. Springer-Verlag.

....must always consist of a complete scan of its local heap. In a special operating mode the creation of a remote reference has to be accompanied by an access to the referenced node [Rudalics, 1986] A modification of the generation scavenging used for Berkeley Smalltalk [Ungar, 1984] was given by Schelvis and Bledoeg [1988] for a distributed Smalltalk collector. In addition to OS, NS, PS and FS which hold cells according to their age, there is additional subspace, RS, that contains all replicated cells . RS is like OS, except that it contains the same cells in the same order on every node. Newly created cells are ....

Schelvis M and Bledoeg E (1988) The implementation of a distributed Smalltalk, ECOOP Proceedings, August 1988 LNCS 322.

....or the reflective techniques, but never with both apart from SOM DSOM. Their proxy metaclass architecture has some common features with ours [FD95] By using a solution based on Smalltalk at the system level, we studied the distributed Smalltalk solutions. Decouchant [Dec86] Schelvis et al. [SB88], McCullough [McC87] and Bennett [Ben87] discussed about the use of proxies in Smalltalk to access remote objects. The two first solutions are based on the Smalltalk VM (virtual machine) and the two other solutions are based on the Smalltalk doesNotUnderstand: method. Indeed, messages for the ....

Marcel Schelvis and Eddy Bledoeg. The Implementation of a Distributed Smalltalk. In ECOOP'88 Proceedings, 1988.

....from other aspects, e.g. the distributed communication technology [6] As seen before, existing technology offer a specific naming system with a well known set of naming properties, targeted to a given domain problem. Naming in distributed object oriented systems normally supports UUIDs, such as[10, 11, 12], whereas naming in most of non object oriented systems supports a hierarchical structured references. Using the Naming pattern application developers can customize and test several naming policies and choose the most suitable for the application needs. Moreover, naming policies are independent of ....

Marcel Schelvis and Eddy Bledoeg. The Implementation of a Distributed Smalltalk. In ECOOP '88, pages 212--232, Oslo, August 1988.

....in tables, can safely be discarded. This does not impose the use of any particular comprehensive GGD policy, any graph tracing based approach can be used. See Louboutin and Cahill [LC95] for further details about an adaptation of an algorithm for GGD inspired by that of Schelvis [Sch89, SB88] using this lazy per cluster log keeping mechanism. Garbage Objects An object becomes garbage when: ffl it is not reachable from any local root, ffl there is no entry in the log of any local cluster associating the object s ID with the ID of some non local cluster. Garbage Clusters A ....

Marcel Schelvis and Eddy Bledoeg. The implementation of a distributed smalltalk. In ECOOP'88, pages 212--232, 1988.

....virtual addresses) to the objects holding the UUID. UUID Name Space UUID Context Objects NameSpace A NameSpace B NameSpace C Context A Context B Context C Application A Application B Application C Figure 3: UUIDs in Object Systems This is the traditional approach in distributed systems, such as [6, 7, 8], in persistent systems, such as [9, 10, 11] and in both distributed and persistent systems [12, 13] Notice that since UUIDs are also used for identification purposes, each object must be associated with one UUID. In persistent systems, where object migration is not supported, the resolution ....

Marcel Schelvis and Eddy Bledoeg. The Implementation of a Distributed Smalltalk. In ECOOP '88, pages 212--232, Oslo, August 1988.

....progressing the collection in the available parts of the system. Whereas each local collector is able to collect local garbage while the rest of the system is unavailable. This further adds efficiency and expedience to the scheme, as most objects tend to be short lived and local [Lieberman 83, Schelvis 88, Jul 88b, Rudalics 86] 5 Comprehensive Garbage Collection In terms of a graph, a comprehensive collection must partition the distributed graph of objects, connected by references, in two very well defined parts. One containing exactly all the objects reachable from the distributed root set, ....

Marcel Schelvis and Eddy Bledoeg. The implementation of Distributed Smalltalk. In S. Gjessing and K. Nygaard, editors, ECOOP'88, European Conference on Object-Oriented Programming, Proceedings published in: Lecture Notes in Computer Science 322, pages 212-- 232, Springer-Verlag, Oslo, Norway, August 1988.

....halt. Local garbage collection causes all activities on that processor to stop. Hence this algorithm is pseudo realtime. Furthermore, interprocessor communication does not involve housekeeping messages. This is handled individually by each processor. DistributedSmalltalk The DistributedSmalltalk [SB88] system consists of cooperating Smalltalk virtual machines distributed over a network that provides complete distribution transparency to the image level. There is a loose cooperation between hosts. Hosts are free to do a local garbage collection anytime. No remote hosts are involved during a ....

Marcel Schelvis and Eddy Bledoeg. The implementation of distributed smalltalk. In Proceedings of ECOOP 88, 1988.

....collection schemes from 1986 1992 (part 2) 37 2.4 Distributed Collection In distributed systems inter node references complicates garbage collection, especially when some nodes are temporarily unreachable. However, many references tend to be short lived and local [Lieberman 83, Schelvis 88, Jul 88b, Rudalics 86] Thus partitioning the storage in node local areas and applying garbage collection locally to each, should collect a large part of the garbage. Most distributed collectors, whether based on local area collectors or not, are satisfied with the collection of most of the ....

....call this the InTable. The global collector removes globally detected garbage from these tables, thus leaving the reclamation to the local collectors. In general, this may be done by removing the global garbage from the local root set. Another version of Distributed Smalltalk by Marcel Schelvis [Schelvis 88, Schelvis 89] is based on local collectors which cooperate by exchanging messages with timestamps. Thus the basic algorithm may be either traversing or counting. Barbara Liskov and Rivka Ladin describes in [Liskov 86] a collection scheme, where independent local collectors work independently ....

[Article contains additional citation context not shown here]

Marcel Schelvis and Eddy Bledoeg. The implementation of Distributed Smalltalk. In S. Gjessing and K. Nygaard, editors, ECOOP'88, European Conference on Object-Oriented Programming, Proceedings published in: Lecture Notes in Computer Science 322, pages 212--232, Springer-Verlag, Oslo, Norway, August 1988.

No context found.

M. Schelvis and E. Bledoeg. The implementation of a distributed Smalltalk. Lecture Notes in Computer Science, 322:212--232, 1988.

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