P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The design and implementation of the clouds distributed operating system. Computing Systems, 3(1), 1990.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....developments in operating systems as well as in programming languages and databases. One emerging model for the support of cooperative distributed applications is that of a distributed shared universe organized as a set of passive objects (active agents are define outside of objects) Bal 92] [Dasgupta 90], Liskov 92] In this paper we report on our experience in designing, implementing, and using a system to support such a model. Our goal is to provide an efficient platform for a family of object oriented languages such as Guide (a language designed by our group [Krakowiak 90] and a persistent ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, U. Ananthanarayanan, U. Ramachandran, M. Ahamad, R.J. LeBlanc, W.F. Appelbe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, C.J. Wilkenloh. The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, vol. 3, 1, pp. 11-46

....For the purpose of this analysis, an activity is defined to be an abstract thread of control that starts out in a process and moves in and out of user level servers and the operating system as calls are made to those servers and the system. This is similar to threads traversing objects in Clouds [26]. Such an abstract thread model corresponds to a synchronous programming style, which is in contrast to an asynchronous programming style where activities 47 are essentially forked by making asynchronous service requests to external servers or kernel primitives. For example, consider an ....

P. Dasgupta et al. The Design and Implementation of the Clouds Distributed Operating System. Computing Systems, 3(1):11--45, Winter 1990.

....developments in operating systems, programming languages and databases. One emerging model for the support of cooperative distributed applications is that of a distributed shared universe organized as a set of objects shared by concurrent activities. designed and implemented (Gothic [ Clouds [Dasgupta90] Emerald [Hutchinson87] Argus [Liskov85] few have addressed the development of protected applications in such an environment (Melampus [ Luniewski91] Birlix [Kowalski90 ] In the Guide project, we have designed and implemented a distributed system that not only intends to support the ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, R. Ananthanarayanan , U. Ramachandran, M. Ahamad, R.J. LeBlanc, W.F. Appelbe , , P.W. Hutto, M.Y.A. Khalidi and C.J. Wilkenloh, The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, 3(1), pp. , Winter 1990.

....as shown in Figure 2. It is a distributed computation which transparently and reliably spans physical nodes, contrary to how conventional threads (conceived as lightweight processes) are confined to a single address space in most other recent OS s such as Mach [48] and Chorus [49] however, Clouds [50] employs a thread model similar to Alpha s. An a thread carries parameters and other attributes related to the nature, state, and service re Object 1 Object 2 Object 3 Object 4 a Thread a a Thread b a Thread c Figure 2: Alpha s Distributed Threads (a Threads) quirements of the ....

Dasgupta, P., R.C. Chen, S. Menon, M.P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R.J. Leblanc, W.F. Appelbe, J.M. Berabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wilkenloh, The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, USENIX, June 1988.

....model for the support of cooperative distributed applications is that of a distributed shared universe organized as a set of objects shared by concurrent activities. While many experimental distributed object oriented systems have been designed and implemented (Gothic [ Bantre91] Clouds [Dasgupta90] , Emerald [Hutchinson87] Argus [Liskov85] few have addressed the development of protected applications in such an environment (Melampus [ Luniewski91] Birlix [Kowalski90 ] In the Guide project, we have designed and implemented a distributed system that not only intends to support the ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, R. Ananthanarayanan , U. Ramachandran, M. Ahamad, R.J. LeBlanc, W.F. Appelbe , J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi and C.J. Wilkenloh, The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, 3(1), pp. 11-45, Winter 1990.

....part of the operating system, which complicates object management. In Emerald [4] a successor project to Eden, static type checking was introduced, and objects may be active or passive. The emphasis on language support is an important feature of Guide, and is also present in Emerald, Clouds [5] and Argus [6] A strong point of Argus is transaction support and recovery mechanisms. A major difference between Argus and Guide is the granularity of objects. While guardians (in Argus) are usually large units, such as servers, objects in Guide are usually small (like files in traditional ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, U. Ananthanarayanan, U. Ramachandran, M. Ahamad, R.J. LeBlanc, W.F. Appelbe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, C.J. Wilkenloh. The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, vol. 3, 1, pp. 11-46

....systems (e.g Emerald [Black86] Orca [Bal87 ] Generic object model The Guide 2 system provides a generic object model for the support of several object oriented programming languages (OOPLs) Several projects attempted to provide such a generic platform. Some earlier systems, such as Clouds [Dasgupta90] or COOL v1 [ Habert90] tried to directly map OOPLs on abstractions provided by some system kernels, essentially address spaces or segments. Since OOPLs deal with fine grained objects, these prototypes either provided two kinds of objects, local objects managed by the compilers and global system ....

....the support of unsafe languages required protection mechanisms at the system level. Therefore, the Guide 2 system maintains a strong isolation between execution structures and objects. Its design may be compared to issues employed in similar projects. Systems such as Argus [Liskov85] or Clouds [Dasgupta90] enforce isolation for coarse grained objects, each context being associated with a unique object. Since local objects are managed at the language level, we may compare their objects to our clusters. As previously noted (section 3.1.2) the major difference is that Guide 2 manages a global naming ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, R. Ananthanarayanan , U. Ramachandran, M. Ahamad, R.J. LeBlanc, W.F. Appelbe , J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi and C.J. Wilkenloh, The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, 3(1), pp. 11-45, Winter 1990.

....of distributed object oriented operating systems. 1. Introduction Several approaches have been considered in projects aiming at implementing a distributed object oriented operating system. Some of them chose to build the entire system from scratch i.e. on a bare machine (e.g. Clouds [Dasgupta 90]) but most of them have chosen to implement a layer on top of the Unix system (e.g. Emerald [Black 86] Argus [Liskov 87] For some years a number of research groups have been experimenting the emerging microkernel technology mainly Mach [Acetta 86] and Chorus[Rozier 88] for building such ....

....an object can only affect objects having the same owner. A truly object oriented protected scheme would have required a separate task for each object, in order to prevent an error in an object from affecting another object, as it is the case in a real segmented based operating systems like Clouds [Dasgupta 90]. This solution is not efficiently applicable here, because the average object size is small (see section 5) Figure 2 illustrates the structure of jobs including protection aspects. O1 A1 A2 O3 O5 O4 06 O8 O9 NODE1 NODE2 NODE3 OWNER2 OWNER3 JOB OWNER1 Fig. 2: Jobs and protection ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahmad, R.J. LeBlanc, W.F. Appelbe, J.M. BernabeuAuban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wilkenloh, The Design and Implementation of the Clouds Distributed Operating System, In Computing Systems , vol. 3, n 1, Winter1990, pp. 11-45

....may have any number (including zero) of loci executing within it. Parameters consisting of data and capabilities for other containers (see below) may be passed on an invoke operation. This mechanism is very similar to those provided in other object based systems such as Monads [14] Clouds [6] , Alpha [5] and Spring [9] One of the key features to the invocation mechanism is that it is potentially very fast in comparison to message passing systems[2, 10] and for this reason a variant of this model has been incorporated into Mach[8] 2.3 Mapping and address space composition The ....

Dasgupta, P., et al "The Design and Implementation of the Clouds Distributed Operating System", Computing Systems Journal, vol 3, pp. 11-46, 1990.

....the introduction of an extra layer of software in this manner can only have an adverse effect on performance. There are numerous examples of operating systems that support persistence in one form or another. Examples include Multics [5, 13, 37] MONADS [26, 39, 40] Eumul L3 [28] Clouds [14], Choices [9] KeyKOS [8, 24] and Grasshopper [17, 41] For the past five years, we have been involved in the design and implementation of the Grasshopper system, an overview of which is presented below. 3. Grasshopper Grasshopper is an operating system explicitly designed to support orthogonal ....

P. Dasgupta, R.C. Chen, et al., "The Design and Implementation of the Clouds Distributed Operating System", Computing Systems, 3(1), pp. 11-46, 1990.

....Naturally, the introduction of an extra layer of software in this manner can only have an adverse effect on performance. There are numerous examples of operating systems that support persistence in one form or another. Examples include Multics [2, 35] MONADS [24, 38, 39] Eumul L3 [27] Clouds [14], Choices [9] KeyKOS [20] and Grasshopper [40] Each of these persistent operating systems attempts to address some or all of the issues described above. 2.# Existing Persistent Operating Systems This Section investigates several different operating systems that provide support for ....

....on which it is based. The hardware design continues to age and hence suffers from poor performance and reliability compared to the modern platforms of today. This limits its usefulness and makes it difficult for others to benefit from the results of the research. 2. 3# Clouds The Clouds system [14, 15] was developed at the Georgia Institute of Technology in the mid 1980 s as a research project aimed at exploring the realms of distributed computing. Clouds is a distributed operating system designed to run without other support on bare hardware. The first version of Clouds ran on VAX hardware and ....

P. Dasgupta, R. Chen, et al., "The Design and Implementation of the Clouds Distributed Operating System" 88/25, Georgia Institute of Technology, 1988.

....persistent application systems on top of conventional operating systems such as Unix, Mach, or Windows NT is prone to inefficiency [25] There are numerous examples of operating systems that support persistence in one form or another. Examples include MONADS [53] 35] 52] Eumul L3 [37] Clouds [18], Choices [11] KeyKOS [30] and Grasshopper [20] Each of these persistent operating systems attempts to address some or all of the issues described above. This Section investigates several of these operating systems. In each case, the abstractions relevant to the implementation of persistent ....

....both persistent. The stability of segments is provided through the virtual memory paging mechanisms in conjunction with a checkpoint mechanism. In the absence of failure, this enables the system to be restarted using the consistent state produced at the last checkpoint. Clouds The Clouds system [18, 19] was developed at the Georgia Institute of Technology in the mid 1980 s. The first version of Clouds ran on VAX hardware and consisted of a monolithic kernel written in C. This version was subsequently abandoned due to its complexity and replaced with an improved design which benefited from the ....

P. Dasgupta, R. Chen, et al., "The Design and Implementation of the Clouds Distributed Operating System" 88/25, Georgia Institute of Technology, 1988.

....Because Opal was not based on a type safe language, resource allocation and reclamation was coarse grained, and based on reference counting of segments. Many research projects have explored operating systems issues within the context of programming languages. For example, Argus [33] and Clouds [16] explored the use of transactions within distributed programming languages. Other important systems that studied issues of distribution include Eden [3] Emerald [11] and Amber [12] These systems explored the concepts underlying object migration, but did not investigate resource management. ....

P. Dasgupta et al. The design and implementation of the Clouds distributed operating system. Computing Systems, 3(1), Winter 1990.

....false sharing. Likewise, data movement under scope consistency is similar to the auto lock mechanism discussed in Section 3.2. 7. Related work Software distributed shared memory (SDSM) has been an active field of research for over a decade. Early page based systems, such as Ivy [24] and Clouds [10], established the basic ideas of page based VM support. Relaxed consistency models were introduced by Munin [6] which pioneered the use of release consistency and multiple protocols, and Treadmarks [20] the first highly portable implementation of lazy release consistency and the system on which ....

P. Dasgupta, R. C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. L. Jr., W. Applebe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wileknloh, "The Design and Implementation of the Clouds Distributed Operating System," in Computing Systems Journal, Winter 1990.

....i.e. to define passive objects being executed by independently defined processes. We did not find strong logical arguments in favour of either solution. Both have been adopted by existing object based systems (e.g. active objects in Argus [19] Eden [5] and Emerald; passive objects in Clouds [11], Amoeba [22] and SOS [26] The choice is mostly influenced by considerations of efficiency and adequacy for the hardware and application domain. The applications that we contemplate involve creating many (usually small) objects, 4 and building large compound structures out of object components ....

P. Dasgupta P. et al., The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, Vol. 3, Num. 1, pp 11-46, (1990).

....developments in operating systems as well as in programming languages. In the 80 s, a model has emerged for the support of cooperative distributed applications, that of a distributed shared universe organized as a set of objects. Distributed objectoriented systems such as Emerald Black87] Clouds [Dasgupta90] or Guide [Hagimont94] belong to this class of systems. More recently, the growth of the Internet, which is now used daily for cooperation, logically leads to the deployment of distributed cooperative applications over the Internet. Today, distributing applications on the Internet is closely ....

....with a DSM, object sharing between heterogeneous sites with dynamic deployment of applications is much more difficult and it is not addressed in the framework of the Perdis project. Our work on Javanaise can also be related to another object based system of the 80s which is the Clouds system [Dasgupta90]. Clouds provides essentially the same paradigm as Guide, i.e. a distributed shared object space, but it differs by the management of object granularity. Clouds objects are coarse grained since an object is implemented by a virtual address space. Clouds objects are developed using the CC (Clouds ....

P. Dasgupta, R. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc, W. Appelbe, J. Bernabeu-Auban, P. Hutto, M. Khalidi, C. Wilkenloh, "The Design and Implementation of the Clouds Distributed Operating System", Computing Systems, 3(1), pp. 11-45, Winter 1990.

.... grained address spaces and new coherence protocols and synchronization primitives [2] 5] 8] Focusing on the former, systems like Mirage[8] and Mirage [9] support sharing of smaller entities called segments (that can implement individual objects) Similarly, systems like Emerald[15] and Clouds[7] directly support sharing of remote objects. Because of problems with false sharing[14] in DSM systems, the trend has been to share smaller entities, down to the level of a single C object. This research has led to a reduction in overhead allowing even moderate grained parallel applications to ....

....Coherence Protocol (DBR) these have discussed in a variety of technical articles [24] 25] 26] 27] Solutions that provide transparent fault tolerance and high data availability for DSM should apply equally well to VDOs. Furthermore, ongoing research in reliable distributed object systems[7][15] could benefit from DSM methods. In short, the synergy between distributed object systems and DSMs systems allows a melding that would benefit both approaches. 6. New Research Directions. Although merging DSM and a distributed object system is a conceptually simple idea, the combination ....

P. Dasgupta, R. Chen, S. Menon, M. Pearson, et. al., The design and implementation of the Clouds distributed operating system, Computing Systems Journal, 3 Winter 1990. A CASE FOR VIRTUAL DISTRIBUTED OBJECTS 9

....balancing with migration. Early research efforts solved these problems by implementing a distributed operating system tailored for a specific hardware platform; examples of such operating systems include V [162] Nest [2] Butler [130] Sprite [44] Amoeba [123] Amber [32] Eden [14] and Clouds [41], Accent [142] 20 Charlotte [58] and Locus [140] More recent research efforts have implemented specific functionality on top of the vendor supplied operating system running on each machine in the distributed system. Many of these run time environments are still in use today and some systems ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The design and implementation of the clouds distributed operating system. Computing Systems, 3(1), 1990.

....in operating systems as well as in programming languages. In the 8Os, a model that emerged for the support of cooperative distributed applications is that of a distributed shared universe organized as a set of objects. Distributed object oriented systems such as Emerald (Black, 1987) Clouds (Dasgupta, 1990) or Guide (Hagimont, 1994) belong to this class of systems. More recently, the growth of the Internet, which is now daily used as a cooperation support, logically leads to consider the deployment of distributed cooperative applications over the Internet. Today, distributing applications on the ....

....using the set of tuple management primitives. We believe that application level objects should be implicitely shared between applications, as in the Corba (or RMI) philisophy. Most of the work described in this paper was influenced by the ideas developed in several projects from the 80 s ((Dasgupta, 1990), Hagimont, 1994) addressing the issue of the design and implementation of distributed object based operating systems. Javanaise can be seen as a grant child of these systems. Its main contribution is to study the application of a similar paradigm (a distributed shared object space) to ....

Dasgupta P., Chen R., Menon S., Pearson M., Ananthanarayanan R., Ramachandran U. , Ahamad M., LeBlanc R. , Appelbe W. , BernabeuAuban J. , Hutto P. , Khalidi M. and Wilkenloh C. (1990) The Design and Implementation of the Clouds Distributed Operating System, Computing Systems, 3(1), pp. 11---45.

....to the protected domain. The unprotected objects are accessible from the unprotected domain, and thus can be passed to protected methods by pointer reference. Acknowledgements The original notions for SPACE were inspired by the ideas about associating general semantics with segments in Clouds Ra[8]. Clouds also helped inspire the name SPACE ) Domains were developed as a solution to object copying in Clouds. The use of virtual memory to provide capabilities is similar to the implementation of the MCP (O S) on the segmented architecture of the Burroughs B5500 6700, and from ideas developed ....

P. Dasgupta, et. al., "The Design and Implementation of the Clouds Distributed Operating System," Technical Report, School of Information and Computer Science, Georgia Institute of Technology.

....in operating systems as well as in programming languages. In the 8Os, a model that emerged for the support of cooperative distributed applications is that of a distributed shared universe organized as a set of objects. Distributed object oriented systems such as Emerald [Black87] Clouds [Dasgupta90] or Guide [Hagimont94] belong to this class of systems. More recently, the growth of the Internet, which is now daily used as a cooperation support, logically leads to consider the deployment of distributed cooperative applications over the Internet. Today, distributing applications on the ....

....which targets distributed applications on the Internet, relies on clusters caching on the client nodes and its clustering policy is implicitly derived from the applications structures. Our work on Javanaise can also be related to another object based system of the 80s which is the Clouds system [Dasgupta90]. Clouds provides essentially the same paradigm than Guide, i.e. a distributed shared object space, but it differs by the management of object granularities. Clouds objects are coarse grained since an object is 16 implemented by a virtual address space. Clouds objects are developed using the ....

P. Dasgupta, R. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc, W. Appelbe, J. Bernabeu-Auban, P. Hutto, M. Khalidi, C. Wilkenloh, "The Design and Implementation of the Clouds Distributed Operating System", Computing Systems, 3(1), pp. 11-45, Winter 1990.

....before it can modify the data. Upon a write miss, a message is transmitted to invalidate all other copies of the data. The thread that generated the miss blocks until all invalidation messages are acknowledged. This single owner consistency protocol is typical of what existing SM systems provide [14, 18, 28]. 2.3 The Write Shared Protocol The write shared protocol allows two or more threads to concurrently modify shared data, even if the modifications occur within the same page of memory. No communication occurs until the modifications are merged at the time of a release, in keeping with the ....

.... RPC 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 0 2 4 6 8 10 12 14 16 Figure 23 Effect of Function Shipping on Quicksort further illustrates the value of Munin s write shared protocol for dealing with false sharing. 9 Related Work We restrict the discussion of related work to software DSM systems [2, 6, 12, 14, 18, 28] and multiprocessors employing relaxed memory consistency models [26] We adapted from the DASH project [26] the concept of release consistency [20] Unlike DASH, which pipelines updates, Munin buffers updates until release time, focusing not only on masking latency, but also on reducing the ....

[Article contains additional citation context not shown here]

P. Dasgupta, R.C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc Jr., W. Applebe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wileknloh. The design and implementation of the Clouds distributed operating system. Computing Systems Journal, 3, Winter 1990.

....state into another consistent system state, regardless of concurrent executions of other transactions and crashes. Despite the recognition given to the concept, however, OS researchers have refrained from adopting transactions in practice. Some exceptions such as Argus [10] Eden [1] and Clouds [6] only prove the rule. The main advantage of programs encapsulated in transactions is their simple structure. Such programs do not have to deal with interleaving of other concurrent programs or worry about system crashes. In other words, the programmer can concentrate on the program per se, even if ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R.J. LeBlanc, W.F. Appelbe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wilkenloh. The design and implementation of the clouds distributed operating system. Computing Systems, 3(1):11--46, Winter 1990.

....resides in a persistent virtual address space which contains both its data and code. Objects are passive, they do not contain a process or thread. They respond to invocations, which result from a thread entering the object to execute an operation [36] Clouds implements a local segmented machine [35]. An object is a set of segments that can be shared across machines. These segments are virtual memory segments and are always assigned the same virtual address [77] Objects are global and have a system name that is unique over the entire distributed system. They are accessible from all compute ....

P. Dasgupta, R. Chen, S. Menon, M. Person, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc, W. Applebe, J. Barnabeu-Auban, P. Hutto, M. Khalidi, and C. Wilkenloh. The Design and Implementation of the Clouds Distributed Operating System. Computing Systems, 3(1), 1990.

....specificity allows the Cilk NOW design to take an endto end approach [38] that leverages properties of the Cilk programming model in order to implement adaptive parallelism and fault tolerance simply and efficiently. Distributed operating systems [17, 36, 43, 46] and remote execution facilities [18, 19, 31, 35, 47] provide services such as remote process execution and, in some cases, process migration. These systems are not intended to be parallel programming environments, though presumably a parallel programming environment could by built atop one of these systems. In fact, Orca [42] which has been built ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The design and implementation of the Clouds distributed operating system. Computing Systems, 3(1):11--46, 1990.

....to continued growth is the difficulty of programming in the message passing paradigm that is native to the hardware. As one way to overcome this obstacle, researchers have developed software distributed shared memory (DSM) systems that enable processes on different workstations to share memory [20, 9, 7, 6, 15, 27, 16]. Despite the recent improvements in memory consistency algorithms [15, 27] hand tuned message passing programs frequently outperform comparable shared memory programs by a significant margin. We believe that compilation techniques developed for data parallel languages, such as regular section ....

....of the benefits of aggregating data communication, especially on a network of workstations. In addition, computation and communication are overlapped. Midway [27] attempts to provide a similar functionality to cross synchronization prefetching at the expense of additional user complexity. Clouds [7] is a distributed system that uses objects to achieve a similar purpose. The disadvantage of these approaches is that data must be explicitly associated by the program with the synchronization variable or object in question, and additional communication is required to change this association. Our ....

P. Dasgupta, R.C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc Jr., W. Applebe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wileknloh. The design and implementation of the Clouds distributed operating system. Computing Systems Journal, 3, Winter 1990.

....may not be appropriate for a channel administration entity, which may have multiple operations pending simultaneously. Because of this need for asynchrony, a message passing paradigm is the best suited for this situation. We note that in an object oriented distributed system such as Clouds [Dasgupta90], some sort of object migration, where channels are represented by objects, may be appropriate. Not having such an environment available to us, we did not investigate this possibility further. 4.0 Decomposition We now describe the decomposition of functionality used in the design of the first ....

P. Dasgupta, R. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc, W. Appelbe, J. Bernabéu-Aubán, P. Hutto, M. Khalidi, and C. Wilkenloh. "The Design and Implementation of the Clouds Distributed Operating System", USENIX Computing Systems, Vol. 3, No. 1, Winter 1990.

....Ideally, the node to be used for executing a given method must be chosen so as to reduce the two conflicting figures that are the number of messages exchanged between processors, and the processors load. Considering existing distributed object based runtime systems (e.g. Argus [2] Clouds [10], Doom [5] Emerald [3] Guide [11] placement of method executions relies on a simple strategy that is one of the following: execution on the storing node of the called object, execution on the calling node, or execution on the node explicitly mentioned by the programmer. Obviously, the two ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanrayanan, R. J. LeBlanc, M. Ahamad, U. Ramachandra, W. F. Appelbe, J. M. BernabeuAuban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenoh. The design and implementation of the Clouds distributed operating system. Computing Systems, 4(3):243--275, 1990.

....object based DBMS was available. Moreover, object based DBMS were expected to be too slow for creating high bandwidth collaborative user interfaces since they translate database references to memory addresses in software. 2 Object models supported by operating systems such as Eden [1] and Clouds [7] met our requirements best. They support shared, distributed, persistent, and protected objects. Moreover, they map objects to virtual memory, thereby letting the hardware take care of address translation. However, current object based operating systems have taken the revolutionary approach of ....

....conventional systems. Overall, our preliminary experience with it has been positive. Several properties of Suite are present in other object based systems. For instance, persistent and multi user objects are also supported by several other systems including Hydra [26] Argus [15] Eden [1] Clouds [7], and GemStone [16] distributed objects communicating via remote procedure calls are supported by most modern systems including Mach [14] Andrew [21] Hermes [5] Argus, Clouds, and Eden; protected objects are supported by ################## 10 Unfortunately, the tool polls the status of ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh, The Design and Implementation of the Clouds Distributed Operating System , Usenix Computing Systems 3(1) pp. 11-46 (Winter 1990).

....The name of the system expresses this goal, since it is an acronym for a System of Uniform Interactive Type directed Editors . In order to build the system, we need the capabilities of remote procedure call [10] and active persistent data [9] While some research had been done in these areas [1, 3], none of the existing systems are compatible with conventional operating systems supporting multiple conventional programming languages. Therefore, we have (somewhat grudgingly) undertaken the task of designing and implementing a software suite that also provides these capabilities. We have ....

....default presentation of data structures, automatic translation between persistent and internal representation of data, and a user interface allowing user control over the degree of feedback. In this report, we highlight mainly those aspects of Suite that are unique to it. Several systems such as [1, 3] call persistent applications communicating with each other objects , and the system on the whole object based . In other papers on Suite [4] we also have used the object based terminology, but in this paper we do not to highlight the fact that unlike other object based systems Suite is an ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh, "The Design and Implementation of the Clouds Distributed Operating System ," Usenix Computing Systems 3:1 (Winter 1990), pp. 11-46.

.... difficult to program links, so created won programming language to control them at a higher level of abstraction. ffl Drawbacks requires existing applications to be rewritten in their language programming model is complex Clouds ffl An implementation of distributed shared memory. [10] ffl Model test bed distributed OS; object oriented capability based system; all programs, devices, data, etc are objects; file system subsumed into object system; all objects have system wide name in flat namespace; nameserver translates user names to sysnames; inter object procedure ....

Dasgupta, P., R. Chen, S. Menon, M. Pearson, R. Ananthanarayanon, U. Ramachandran, M. Ahamad, R. LeBlanc, W. Applebe, J. Bernabeu-Auban, P. Hutto, M. Khalidi, and C. Wilkenloh, "The Design and Implementation of the Clouds Distributed Operating System," Computing Systems, v.3, No. 1, Winter 1990.

....balancing with migration. Early research efforts solved these problems by implementing a distributed operating systems tailored for a specific hardware platform; Examples of such operating systems include V [156] Nest [2] Butler [125] Sprite [40] Amoeba [118] Amber [29] Eden [11] and Clouds [37], Accent [137] 53] and Locus [135] More recent research efforts have implemented specific functionality on top of the vendor supplied operating system running on each machine in the distributed system. Many of these run time environments are still in use today across a variety of platforms and ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The design and implementation of the clouds distributed operating system. Computing Systems, 3(1), 1990.

....differs in that we take into account the objects characteristics in addition to the load of the system s nodes, in the placement decision. Considering existing distributed object based runtime systems, placement of method executions relies on a simple strategy that is one of the following [14, 15, 16]: execution on the storing node of the called object, execution on the calling node, or execution on a node explicitly mentioned by the programmer. Obviously, the two first (transparent) policies are not satisfying since they do not exploit the processors load. On the other hand, the approach ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanrayanan, R. J. LeBlanc, M. Ahamad, U. Ramachandra, W. F. Appelbe, J. M. Bernab'euAuban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenoh, "The design and implementation of the Clouds distributed operating system," Computing Systems, vol. 4, no. 3, pp. 243--275, 1990.

....of windows. Finally, it provides a flexible implementation model that allows programmers to extend the coupling model. 6.3. Coupling Arbitrary User Interfaces Several systems support communication of information among agents managing different user interfaces. Argus [21] Eden [1] and Clouds [6] allow multiuser applications to be constructed as distributed objects that communicate with each other using high level remote procedure calls. Field [28] and Mediators [35] provide methods for easily changing the communication patterns among communicating objects, thereby supporting ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh, "The Design and Implementation of the Clouds Distributed Operating System ," Usenix Computing Systems 3:1 (Winter 1990), pp. 11-46.

....a thread attempts to write to replicated data, a message is transmitted to invalidate all other copies of the data. The thread that generated the miss blocks until all invalidation messages are acknowledged. This single owner consistency protocol is typical of what existing DSM systems provide [21, 27, 42], and is what we use exclusively to represent a conventional DSM system in our performance evaluation. Once read only data has been initialized, no further updates occur. Thus, the consistency protocol simply consists of replication on demand. A runtime error is generated if a thread attempts 1 ....

....to false sharing. It is up to the programmer or the compiler to lay out the program data structures in the shared address space such that false sharing is reduced. The directory management scheme in our implementation is largely borrowed from Ivy s dynamic distributed manager scheme. Both Clouds [21] and Mirage [27] allow part of shared memory to be locked down at a particular processor. In Clouds, the programmer can request that a segment of shared memory be locked on a processor. In Mirage, a page remains at a processor for a certain Delta time window after it is modified by that ....

[Article contains additional citation context not shown here]

P. Dasgupta, R.C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc Jr., W. Applebe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wileknloh. The design and implementation of the Clouds distributed operating system. Computing Systems Journal, 3, Winter 1990.

....Therefore a parallel paradigm must be provided that enables the user to take complete control over mapping and communication of tasks. In MuPAD there exist two different kinds of parallelism micro and macro parallelism similar to CLIDE [22] Distributed Eiffel [21] both based on Clouds [4]) or LADY [19] An overview over most existing parallel computeralgebra systems can be found in [16] 2 Micro Parallelism 2.1 Motivation Many parallel programs use very fine grained parallelism very many very quickly solvable tasks are produced. Using tightly coupled computers it is worth ....

Dasgupta et. al. The design and implementation of clouds distributed operating systems. USENIX Computing Systems Journal, 3(1), 1990.

....(demand paging coherence = DSM) we believe that this is more a tribute to the quality of the work than anything else. Since this work, there has been a huge amount of work done to extend the idea to other areas (e.g. distributed object based systems[14, 5, 1, 20, 10] and operating systems[23, 7, 2]) and to improve its performance[11, 4, 24, 15] However, the impact of This research was supported in part by the National Science Foundation under Grant CCR 9308879 and by the Advanced Research Projects Agency (ARPA) monitored by the Department of the Army, under Contract DABT63 94 C 0058. ....

....article are those of the authors and should not be interpreted as representing official views or policies of the Advanced Projects Research Agency, the U.S. Government, or the University of Utah. DSM on non research users and applications has been very small. Why is that The earliest DSM systems[17, 11, 14, 7] provided the basic functionality of a virtually shared address space spanning a network of machines, but their use of the same coherence protocols as shared memory hardware resulted in poor performance for applications with even a moderate amount of fine grained sharing. The reasons for this were ....

[Article contains additional citation context not shown here]

P. Dasgupta, R.C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc Jr., W. Applebe, J.M. BernabeuAuban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wileknloh. The design and implementation of the Clouds distributed operating system. Computing Systems Journal, 3, Winter 1990.

....One emerging model for the support of cooperative distributed applications is that of a distributed shared universe organized as a set of objects shared by concurrent activities. While many experimental distributed object oriented systems have been designed and implemented (Gothic [2] Clouds [7], Emerald [10] Argus [14] few have addressed the development of protected applications in such an environment (Melampus [15] Birlix [12] In the Guide project, we have designed and implemented a distributed system that not only intends to support the above model, but also provides mechanisms ....

P. Dasgupta, R.C. Chen, S. Menon, M.P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R.J. LeBlanc, W.F. Appelbe, J.M. BernabeuAuban, P.W. Hutto, M.Y.A. Khalidi, C.J. Wilkenloh, "The Design and Implementation of the Clouds Distributed Operating System", Computing Systems, vol. 3, num. 1, pp. 11-45, Winter 1990.

....September 1993 they are generally tied to a particular machine architecture and thus nonportable. Also, current operating systems generally provide more functionality than is needed or wanted for parallel processing (e.g. virtual memory management) Some modern operating systems, like Clouds [12], support their own object model. Unless such a model is very simple, flexible, and lightweight, layering another object model on top of it can be troublesome and inefficient. Higher level approaches to supporting parallel programming include page based Distributed Shared Memory (DSM) systems ....

P. Dasgupta, R.C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc Jr., W. Applebe, J. M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C. J. Wilkenloh. The Design and Implementation of the Clouds Distributed Operating System. Computing Systems Journal, 3, 1990.

....conglomerate of nodes [Bartoli93] This means that in the end, the same methods for sharing data over address space boundaries will be needed as is the case in today s process based systems, albeit for sharing across a WAN or large LAN. 2. 3 Clouds The Clouds distributed operating system, see [Dasgupta90, Dasgupta88] is object based in that all data, programs, devices and other resources are modelled as objects. Objects are the passive elements in the system, whereas threads are the active ones. An object can be viewed as a persistent virtual address space in which threads may start executing at certain entry ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The design and implementation of the Clouds distributed operating system. Computing Systems, 3, 1990.

....SPLASH application suite. Related work in this area are covered in Section 8, and concluding remarks are presented in Section 9. 2 2 A Base Software DSM Several DSM systems have been proposed in the past. Systems such as [6, 17] implement the DSM abstraction in hardware. However, most systems [9, 4, 10, 29, 27, 26, 3] use a model in which a cluster of machines are connected together using a network. On such systems, often referred to as Software DSM, the DSM abstraction is implemented by a software layer. We describe a base software DSM system that has some of the common features of recent software DSM ....

....such applications on a highly available DSM which is part of our future work. 8 Related Work Several DSM systems have been proposed in the past [20] In some cases, the DSM abstraction is implemented in hardware [6, 17, 7] but in most cases it is implemented either entirely in software [10, 9, 4, 5, 12, 29, 3, 26, 27] or in a hybrid hardware software layer [2, 16, 25] One possible problem with a DSM system is that having many processing units collaborating to perform a single task increases the probability of a fail11 ure. Fault tolerance is an issue in DSM systems due to the multiple points of independent ....

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabu-Aubn, P. W. Hutto, M. Yousef, A. Khalidi, and C. J. Wilkenloh. The design and implementation of the clouds distributed operating system. Computing Systems, 3(1):11--46, 1989.

No context found.

P. Dasgupta, R. C. Chen, S. Menon,M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W.Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The Design and Implementation of the Clouds Distributed Operating System. Computing Systems, 3, 1990.

No context found.

P. Dasgupta, R. C. Chen, S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The design and implementation of the clouds distributed operating system. Computing Systems, 3(1), 1990.

No context found.

P. Dasgupta, et al.: "The design and implementation of the Clouds distributed operating system", Computing Systems Journal, 3, Winter 1990.

No context found.

P. Dasgupta, R. C. Chen, , S. Menon, M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. Leblanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W. Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The design and implementation of the Clouds distributed operating system. Computing Systems, 3(1):11--46, Winter 1990.

No context found.

P. Dasgupta and R. C. Chen, et. al., The Design and Implementation of the Clouds Distributed Operating System, Technical Report,(Computing Systems Journal, Vol 3, USENIX, Winter 1990), ftp://helios.cc.gatech.edu/pub/papers/design.ps.Z

No context found.

Dasgupta, Chen, Menon, Pearson, Ananthanarayanan, Ramachandran, Ahamad, LeBlanc, Applebe, Bernebeu-Auban, Hutto, Khalidi, and Wilkenloh. The design and implementation of the Clouds distributed operating system. Computing Systems, 3(1):11--46, 1990.

No context found.

P. Dasgupta, R. C. Chen, S. Menon,M. P. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. J. LeBlanc, W. F. Appelbe, J. M. Bernabeu-Auban, P. W.Hutto, M. Y. A. Khalidi, and C. J. Wilkenloh. The Design and Implementation of the Clouds Distributed Operating System. Computing Systems, 3, 1990.

No context found.

P. Dasgupta, R.C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc Jr., W. Applebe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wileknloh. The design and implementation of the Clouds distributed operating system. Computing Systems Journal, 3, Winter 1990. 116

No context found.

P. Dasgupta, R.C. Chen, S. Menon, M. Pearson, R. Ananthanarayanan, U. Ramachandran, M. Ahamad, R. LeBlanc Jr., W. Applebe, J.M. Bernabeu-Auban, P.W. Hutto, M.Y.A. Khalidi, and C.J. Wileknloh. The design and implementation of the Clouds distributed operating system. Computing Systems Journal, 3, Winter 1990.

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