U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. Coherence of Distributed Shared Memory: Unifying Synchronisation and Data Transfer. In International Conference on Parallel Processing, pages II--160--169, 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....gestion de la coherence entre les copies multiples est faite en gnral au niveau de la page [Bantre 91] par le gestionnake de pagination. Le niveau de coherence entre les copies peut tre strict (comme si tousles sites partageaient une mme mmoire contenant l objet) ou obir h des critres plus lches [Ramachandran 89] Boyer 9lb] Carter 91 ] 2.2.2. Protection et s(curit Les concepts propres h garantir la confidentialit6 et l int6grit6 des informations d un systme se sont d6velopp6s dans les ann6es 60 et ont vu leur premiere mise en oeuvre dans les systmes multi programm6s et les systmes h partage de temps ....

U.Ramachandran, M. Ahamad, Y. A. Khalidi 89, Coherence of Distributed Shared Memory: Unifying Synchronization and data Transfer, International Conference on Parallel Processing, 1989 58

....be classified on the basis of synchronization points in a sequence of shared accesses [Ni91] With Ivy s strict coherence, every read or write is a synchronization point. Munin s release consistency is based on acquire and release operations which are similar to ARCADE s lock and unlock. Clouds [Cn92,Ra89] offers both strict and weak coherency. Weaker coherency typically increases the concurrency of shared data accesses, but their use depends on the application s ability to tolerate stale data. Therefore, application specific coherence policies [An92] can be more efficient. Applications can use ....

Ramachandran, U., Ahamad, M. and Khalidi, M., Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer, Proceedings of the

....on a software implementation of caching at the page size granularity, and he uses a directory based invalidation scheme to ensure data coherence. The work of Francioni, Poplawski, and Pahwa [FPP88] avoids the problem of maintaining memory consistency by caching only program text pages. Clouds [RAK89] is an object oriented system that provides a shared memory abstraction on a network of workstations, but attempts to minimize the number of invalidation messages that result from active write sharing of data. The idea is to integrate the coherence mechanism with synchronization so that the ....

....that all memory references are to local memory. Unlike Am Am N, the Am Ar N policy creates multiple copies of shared pages in preference to moving existing copies. This is essentially the same strategy used by the various distributed virtual memory systems discussed in the literature [Li88, LS89, RAK89, BCZ90b] though the page coherence strategies differ in some cases. The Am Ar N policy performs quite poorly for most of our test applications. When run under the Am Ar N policy, the fish, hough, psolu, hh3d, and gauss applications all experience a page bouncing condition in which shared pages ....

U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. Coherence of distributed shared memory: Unifying synchronization and data transfer. In Proceedings of the 1989 International Conference on Parallel Processing, pages II--160--II169, August 1989.

....in software has been studied by several groups. Kai Li s work [36, 37] based on an ondemand copying of pages between memories (page granularity software caching) with a directorybased invalidation coherency scheme, provides a complete shared memory abstraction to the programmer. Clouds [40] is an object oriented system that provides a shared memory abstraction on a network of workstations, but attempts to minimize the number of invalidation messages that result from active write sharing of data (e.g. synchronization data) The results of this work can be 2 Pronounced ducks ....

U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. Coherence of distributed shared memory: Unifying synchronization and data transfer. In Proceedings of the 1989 International Conference on Parallel Processing, pages II--160--II169, August 1989.

....may be maintained through software, hardware or a combination of the two. The majority of scalable hardware based systems with a general interconnect use invalidations to maintain consistency [10, 21, 9, 13] Several of the software based schemes use a combination of invalidations and updates [2, 14, 11, 3, 24, 23]. This paper presents two hardware controlled update based cache coherence protocols: one based on a centralized directory and the other based on a singly linked distributed directory. The paper considers two major disadvantages of the update based protocols and develops approaches to overcoming ....

U. Ramachandran, M. Ahamad and M. Y.A. Khalidi, "Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfers", Proc. 1989 International Conference on Parallel Processing, pages II160-II169, August 1989.

....can be used to provide a logically shared memory on message passing multicomputers (e.g. the Intel Paragon and the IBM SP1) or on networks of workstations. Examples of software based approaches include Li s Ivy [15] Rice s Munin [4] CarnegieMellon s Mach [10] and Georgia Tech s Clouds [17]. The relative advantages and disadvantages of the hardware based and software based approaches are fairly obvious: hardware based approaches are more costly but can be expected to deliver better performance while software based approaches require less special hardware but may suffer in terms of ....

U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. Coherence of distributed shared memory: Unifying synchronization and data transfer. In Proceedings of the 1989 International Conference on Parallel Processing, pages II-160-II169, August 1989. Page 21 of 21

....to pages that are also shared by other processors, and invalidate or update other copies of the page. Faults to pages currently residing at other processors cause those pages to be copied to the faulting processor. Examples of such systems include Ivy [53] Mermaid [74] Amber [17] Clouds [62], Munin [12] Midway [9] and Treadmarks [44] In Ivy and Mermaid, copies of a page are invalidated as soon as a write to a page is detected by one processor. This makes these systems susceptible to the problem of false sharing at page granularity. The other sys Chapter 2: Review 15 tems take ....

U. Ramachandran, M. Ahamad, and M. Khalidi. Coherence of distributed shared memory: Unifying synchronization and data transfer. In Proc. of the 1989 Conference on Parallel Processing, pages 160-169, June 1989.

....They define a weakly ordered system, where synchronization operations are made explicit to the memory system, and consistency maintenance is done only at synchronization points. One of the earliest implementations where the data transfer is linked with the synchronization operations is reported in [50]. The DASH multiprocessor [24] is a weakly ordered system which implements a memory model called release consistency. Release consistency refined the weakly ordered approach by dividing synchronization operations into two types: acquire and release. Release consistency allows remote memory ....

U. Ramachandran, Mustaque Ahamad, and Y. Khalidi. Coherence of distributed shared memory: Unifying synchronization and data transfer. In Proceedings of the International Conference on Parallel Processing, August 1989.

....Furthermore, after creating a data unit, the ARCADE kernel assumes responsibility for maintaining consistent shared images of the data unit among a network of potentially heterogeneous machines. 3.2 Clouds Clouds is an object based distributed operating system under development at Georgia Tech. [19,20]. The two primary abstractions in Clouds are passive objects and active threads. Objects are similar to abstract data types; they consist of encapsulated code and data which implements defined entry points. Each object occupies a portion of a global virtual address space that spans the entire ....

....associated with page granularity in the general case are not of concern here. Another similarity between Clouds and Li is the fact that both support the notion of one global address space for the entire distributed system. While the Clouds developers find this to be an attractive characteristic [19], it clearly does not scale well for large interconnections. Despite these similarities, There are some significant differences between Clouds and the Li approach. First, the Clouds developers were concerned with the high cost of invalidation messages. Thus, while their DSMC protocol provides as ....

[Article contains additional citation context not shown here]

Ramachandran, U., Ahamad, M. and Khalidi, M., Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer, Proceedings 1989 International Conference on Parallel Processing, Volume II, August, 1989, pp. 160-169.

....test and set. Only one gets the lock. To reduce mis trials before rechecking a lock, an exponential backoff delay after release of a lock with test test and set lessens contention[2] The delay is locally doubled whenever the lock is unlocked, but an attempt to obtain it fails. Queue based locks[9, 10, 2, 15, 16, 3, 18] are alternatives to retested locks. A lock request is sent to a lock owner. If the lock is free, permission is granted; if busy, the request is queued. When the lock is freed, the next queued process gets permission. Lock queues can be supported in hardware[9, 15] or in software[10, 2, 16, 3, ....

....2, 15, 16, 3, 18] are alternatives to retested locks. A lock request is sent to a lock owner. If the lock is free, permission is granted; if busy, the request is queued. When the lock is freed, the next queued process gets permission. Lock queues can be supported in hardware[9, 15] or in software[10, 2, 16, 3, 18]. Test and set with hardware support[20] is adequate on tightly coupled multiprocessors. Queue based locks are needed in distributed memory systems to minimize network traffic after lock release. On multicomputers connected by networks, locating the lock owner is an issue. Distributed directory ....

[Article contains additional citation context not shown here]

U. Ramachandran, M. Ahamad, M. Yousef, and A. Khalidi. Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer. Int. Conf. on Parallel Processing, II:160--169, Aug. 1989.

....strategies are briefly described in section 2.2. 2.1 Design Issues Distributed shared memory provides users a convenient programming abstraction to design their parallel programs. However, the memory consistency between local memories of processors should be maintained. A few literatures [4, 7, 8, 9, 10, 14, 16, 17, 20, 22] address on this problem. In [2, 22] they have shown that no single algorithm for distributed shared memory will be suitable for most applications. The reason is that the shared objects may have different access characteristics. Therefore, when maintaining memory coherence, we should select ....

....approach with the concept of virtual memory in enforcing strong coherence. The drawback of this approach is a lack of consideration for access pattern of shared data. The IVY is more suitable for parallel processing. A hypercube DSM implementation is considered in [15] based on Li s method. Clouds [20] is a distributed operating system being developed at Georgia Tech implemented on a network of SUN 3 workstations. The kernel is known as Ra [3] Ra supports a novel feature, i.e. first DSM system supporting the abstraction of object oriented mechanisms. The combination of memory coherence ....

U. Ramachandran, M. Ahamod, M. Yousef and A. Khalidi, Coherence of distributed shared memory: unifying synchronization and data transfer, Proc. 1989 International Conference on Parallel Processing (Aug. 1989) II-160-II-169.

....a coherence algorithm similar to Li s. Since Mach and Chorus have been ported to several multiprocessor platforms, Mach and Chorus based VSM systems can be potentially available on a variety of platforms. Clouds Clouds [54] is an object based distributed operating system. Ramachandran et al. [157] have proposed the implementation of distributed shared memory in the Clouds kernel. This work is based on Li s work, but provides additional efficient support for synchronisation by integrating synchronisation with cache coherence. In addition to systems described above, many modern (distributed) ....

U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. Coherence of Distributed Shared Memory: Unifying Synchronisation and Data Transfer. In International Conference on Parallel Processing, pages II--160--169, 1989.

....in order to minimize software complexity. It must also be flexible enough to support a wide variety of applications. Furthermore, it must be fast and efficient so that high performance applications can be constructed. Various process interaction paradigms, including RPC [7] object based systems [4, 10, 20], and distributed shared memory [16, 17] have been studied and developed. Unfortunately, each has shortcomings in terms of ease of use, flexibility or performance. This is not surprising, since these requirements are, at best, mutually incompatible. To address this difficult optimization problem, ....

Ramachandran, U., Ahamad, M. and Khalidi, M., Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer, Proceedings 1989 International Conference on Parallel Processing, Volume II, August, 1989, pp. 160-169.

....memory based program, than to a message passing based program. For these reasons and others, much research has been done on how to define and implement distributed shared memory, both from the theoretical point of view [2, 3, 6, 10, 11, 13, 22, 26, 27, 28] and from the practical point of view [14, 17, 18, 19, 21, 31, 32, 33]. However, most of this work ignores the possibility of failures and does not allow processors to join an ongoing computation, in order to speed it up. On the other hand, it is clear that if distributed shared memory is to be implemented in massively parallel computers, or over a network of ....

U. Ramachandran, M. Ahamad, and M. Y. Khalidi. Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer. In Proc. International Conf. on Parallel Processing, pages II--160--169, 1989.

....and Symmetry, and the BBN GP1000 and TC2000 systems; among research prototypes are the NYU Ultracomputer [51] the IBM RP3 [104] and the Illinois Cedar [44] machines. Furthermore, a considerable effort in software development is designed to provide the illusion of shared memory on multicomputers [26, 20, 79, 78, 108, 40, 22]. By restricting our attention to a given class, we filter out some of the strong differences, allowing ourselves to understand the performance within a class more precisely. The execution time of a task on a multiprocessor may be nondeterministic on account of queueing delays due to contention ....

U. Ramachandran, M. Ahamad, and M.Y.A. Khalil. Coherence of distributed shared memory: Unifying synchronization and transfer of data. In Proc. Intl. Conf. on Parallel Processing, volume II, pages 160--169, August 1989.

....to build distributed dynamic data structures. 1. Introduction The distributed shared memory paradigm [1] was originally proposed to simplify the task of writing distributed algorithms. It was felt that applications would be simplified if they did not have to explicitly communicate. Subsequent work [2][3] 4] has focused on improving the efficiency of distributed shared memory and dealing with the problems caused by page granularity. An alternative form of shared data is to use shared data objects [5] In this paper, we discuss three examples of distributed shared memory which support the claim ....

Ramachandran, U., Ahamad, M. and Khalidi, M., Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer, Proceedings 1989 International Conference on Parallel Processing, Volume II, August, 1989, pp. 160-169.

....include speeding up the transfer of a message through a given system by implementing better routing algorithms, and or distributing memory in ways that reduce slowdowns caused by message traffic bottlenecks. Other work has proposed introducing multiprocessor cache memories [YYF85] DSB88] BNR89] [RAK89] and distributing memory [BaJ88] or balancing the message load [ReT86] to limit contention. Another approach would be to ask: Can we achieve better system utilization by increasing the amount of productive activity during global memory references, and reducing the amount of inter process ....

U. Ramachandran, M. Ahamad and M. Khalidi, Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer, in Proc. 1989 International Conference on Parallel Processing, IEEE, 1989, II-160-169.

....The problem of cache coherence in shared memory multiprocessors has been studied almost as long as there have been such machines. A wide range of solutions have been proposed with varying degrees of hardware software support required in the processors, cache controllers, and global memory units [7, 16, 3, 5, 4, 17, 18, 12, 8, 2, 10, 13, 1, 9, 6, 11, 15]. Our interest is in finding a solution with the following desirable properties: 1. The processor hardware need not have any knowledge that shared data is cached in multiple local Present address: Hewlett Packard Laboratories, 1501 Page Mill Road, Palo Alto, California 94304. memories. This ....

....designed to maintain sequential consistency while reducing the performance impact of false sharing. Ivy [17] uses a write invalidate protocol on page size blocks. Since large blocks are more likely to be falsely shared, it is left to the compiler or programmer to effectively align the data. Clouds [18], which uses objects as its unit of consistency, allows the processor to lock the object to avoid cross invalidates. Mirage[12] automatically locks pages, its unit of consistency, for a cer tain amount of time. Orca[4] uses reliable broadcast for both invalidate and update protocols. Amber[8] ....

U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer. In Proc. of the 1989 Conference on Parallel Processing, pages II--160--II--169, June 1989.

....to provide complete DSM functionality, the pager provides two synchronization mechanisms. Clients that need guaranteed continuous access to a set of pages may choose to pin those pages to the local machine. This acts much like the locking protocols supported by other DSM implementations [Li86] [Ra89] [Co92] In addition, support is provided for event notification. Thus, a receiving client can be notified whenever a sending client moves pages into the segment. Notification is based on installable callback routines which are normally defined during segment allocation. These routines are then ....

Ramachandran, U., Ahamad, M. and Khalidi, M., Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer, Proceedings of the 1989 International Conference on Parallel Processing, Volume II, August, 1989, pp. 160-169.

....section. 1.3 Synchronization Hardware primitives for repeated lock tests such as Test and set[3] Test test and set[17] and their extensions[1] evolved on shared memory multiproces sors. In distributed systems repeatedly testing locks produces too much network traffic. Queue based locks[9, 10, 1, 5, 14, 2, 16] are alternatives. A lock request is sent to a lock owner. If the lock is free, permission is granted. If it is busy, the request is queued. When the lock becomes available, the next process in the queue gets permission. Lock queues can be supported in hardware[9, 5] or software[10, 1, 14, 2, 16] ....

....10, 1, 5, 14, 2, 16] are alternatives. A lock request is sent to a lock owner. If the lock is free, permission is granted. If it is busy, the request is queued. When the lock becomes available, the next process in the queue gets permission. Lock queues can be supported in hardware[9, 5] or software[10, 1, 14, 2, 16]. Queue based locks are needed in distributed memory systems, even those with local lock copies, to lessen network traffic after lock release. When moving from multiprocessors connected by busses to multicomputers connected by networks, locating the lock owner becomes an issue. Distributed ....

[Article contains additional citation context not shown here]

U. Ramachandran, M. Ahamad, M. Yousef, and A. Khalidi. Coherence of Distributed Shared Memory: Unifying Synchronization and Data Transfer. International Conference on Parallel Processing, II:160--169, August 1989.

....a mesh interconnect. Rather than storing exact pointer the copy set only tells the probable existence of the copies of a data page in one of the four directions at a mesh point (similar to Li s probable owner concept) Clouds Clouds [29, 30] is an object based distributed operating system. In [73], Ramachandran et al. have proposed implementation of distributed shared memory in the Clouds kernel. This work is based on Li s work but provides additional efficient support for synchronisation by integrating synchronisation with cache coherence. In Clouds, virtual shared memory is used for ....

U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. `Coherence of Distributed Shared Memory: Unifying Synchronisation and Data Transfer'. In International Conference on Parallel Processing, pages II--160--169, 1989.

....of distributed laboratory applications is in exploiting some of the well known database technologies in this domain without any of the usual rigidity of database systems such as serializability. Previous Research. We have undertaken several research projects in areas related to state sharing [53, 54, 55, 38]. One project addressed the implementation of shared memory abstractions on workstation networks. It addressed the formal characterization of shared memory systems [34, 48] and also implemented consistency protocols derived from a number of approaches that have been proposed for building high ....

U. Ramachandran, M. Ahamad, M. Y. A. Khalidi, "Coherence of distributed shared memory: Unifying synchronization and data transfer," In the Proceedings of the 18th International Conference on Parallel Processing, August 1989, Vol II:pp.160-169.

....better performance. However, if different eventcount variables on the same shared page are frequently accessed on different processors, thrashing of the page may result due to false sharing. Clouds Clouds is a DSM system developed at Georgia Tech on a network of workstation connected via Ethernet [26]. It supports an object based model of programming and allows for the migration of objects and other shared data to the processors where they are needed. The object model of computation integrates the synchronization mechanism by providing atomicity in the form of remote procedure calls. An object ....

U. Ramachandran, M. Ahamad, and Y. Khalidi. "Coherence of Distributed Shared Memory Unifying Synchronization and Data Transfer". In International Conference on Parallel Processing, pages 160--169, aug 1989.

No context found.

U. Ramachandran, M. Ahamad, and M. Y. A. Khalidi. Coherence of Distributed Shared Memory: Unifying Synchronisation and Data Transfer. In International Conference on Parallel Processing, pages II--160--169, 1989.

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