A. Itzkovitz, A. Schuster, MultiView and Millipage --- fine-grain sharing in page-based DSMs, in: Proceedings of the 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....re executed at the next synchronization point. references because it is executed after all remote data has been obtained. The basic idea behind the implementation of Implicit 2D is to allow several different views of a single page. This is similar to the work pioneered by the Millipede system [9]. On node i, we create three views for each page using mmap; one is owned by node i and allows reads and writes to complete uninterrupted. The other two are logically owned by nodes i 1 and i 1, assuming that node i is an interior node. These views are protected with PROT NONE, which means ....

A. Itzkovitz and A. Schuster. Multiview and Millipage -- fine-grain sharing in page-based DSMs. In Proceedings of the Third Symposium on Operating Systems Design and Implementation, Feb. 1999.

....of a DSM into small, say 32 to 128 byte chunks, called cache lines (not to be confused with a processor s cache lines) creating a fine grained DSM. There are several hybrids between these three approaches, for example, a mix of a page based and cache line based DSM has been attempted in Multiview [27]. cJVM [3] mixes elements of an object and cache line based DSM by defaulting to RPC but at runtime trying to optimize by using object caching and equating the object to a cache line. By default, Jackal acts as a cache line based DSM. As an optimization, Jackal tries to use function shipping ....

A. Itzkovitz and A. Schuster. MultiView and Millipage - Fine-Grain Sharing in PageBased DSMs. In Proceedings of the 5th USENIX Symposium on Operating Systems Design and Implementation (OSDI '99), pages 215--228, New Orleans, USA, February 1999.

....are exploited to reduce the amount of data transferred to maintain consistency. LOTEC builds indirectly on earlier work on distributed object systems (including [17, 18] and objectbases (including [14, 10] LOTEC is most closely related to work by Fleisch and Hyde [11] and Itzkovitz and Schuster [12]. 3. The Problem and Environment 3.1. The Environment As with LOTEC [20] we assume an environment based on distributed persistent objects. Users interact with the environment by invoking methods on specific objects to accomplish their desired computations. These methods execute as transactions ....

A. Itzkovitz and A. Schuster. MultiView and Millipage:Fine-Grain Sharing in Page-Based DSMs. In Proc. Conference on Operating Systems Design and Implementation, 1999.

....in whichitshould operate. Our work builds indirectly on earlier work on distributed object systems(including [RB94, SF95, BHAB96] and, in some ways, on objectbase systems (including [LLOW91, Deu90] Itismost closely related to work by Fleischand Hyde [FH98] and Itzkovitz and Schuster [IS99] The rest of this paper is organized as follows. Section 2 describes the problem domain and motivates our solution strategy. Section 3 describes nested object transactions and our correctness criterion for them. In Section 4, we present the Lazy Object Transactional Entry Consistency (LOTEC) ....

....startup latencies. This allows us to assess the potential messaging performance of LOTEC under anyofanumber of schemes that reduce the software induced factors in message latency. Theimportance of minimizingmessage latency has recently been illustrated bythe success of the Millipage system [IS99] which exploits Gbps networks and low latency protocols to support fine granularity sequential consistency with performance comparable to existing weak consistency systems. 5.1 Locking Overhead Theoverall performance of a distributed object system built using theLOTEC protocol will be ....

Ayal Itzkovitz and Assaf Schuster. MultiView and Millipage--Fine-Grain Sharing in PageBased DSMs. In Proc. of the Conference on Operating Systems Desin and Implementation (OSDI '99),February 1999.

....on a variety of input data sets and interpret the range of performance levels achieved. The following Chapter 6 develops the final ingredient for a fully automatable parallelization. We show how speculative parallelization can be performed in a languageindependent manner based on the MultiView[IS99] fine grain variable access protection technique. Because we need to audit data accesses at the element level, we propose a more elaborate variable transformation which exposes every element to access protection. We further show that while this technique may induce secondary e#ects on the ....

....6.1: Example of MultiView. 6.4 E#cient Memory Profiling In order to provide an e#cient, language independent, platform independent methodology for the required element by element memory profiling, we turn to recent developments in the Distributed Shared Memory arena. Itzkovitz and Schuster[IS99] pioneered a methodology called MultiView for finegrained memory access control of DSM using the standard OS page fault mechanism. MultiView can support access faults at the variable level, even for variables which share pages with other variables. The way in which this is done is through the ....

[Article contains additional citation context not shown here]

A. Itzkovitz and A. Schuster. Multiview and Millipage---Fine-grain Sharing in Page-based DSMs. In Operating Systems Design and Implementation, February 1999.

.... to provide supercomputing capabilities at a fraction of the cost of traditional multiprocessor systems [1, 8] At the same time, much research has been conducted in software distributed shared memory (DSM) to make it as easy to program clusters as it is to program shared memory multiprocessors [15, 2, 6, 20, 21, 29, 27, 26, 17]. These advances in programmability and performance are making it possible to use very large clusters as a cost e ective platform for data intensive, long running applications. As cluster size and application running times increase, adding fault tolerance becomes critical. At the same time, to ....

A. Itzkovitz, A. Schuster. MultiView and Millipage - Fine-Grain Sharing in Page-Based DSMs. Proc. 3rd USENIX Symposium on Operating Systems Design and Implementation (OSDI), pp. 215-228, February 1999.

....causal coherence, the eventual coherence, etc. The recent LOTEC protocol [24] maintains consistency for nested object transactions. This protocol spares the programmer the burden of explicitly specifying synchronization operations needed for transactional processing. Finally, the novel Millipage [27] technique enables efficient control over the granularity of a shared unit, thus enabling applications to achieve good performance while maintaining sequential consistency. Object based shared memory systems, in which consistency guarantees are given per object, have also been studied. Orca [10] ....

Ayal Itzkovitz and Assaf Schuster. MultiView and Millipage --- Fine-Grain Sharing in Page-Based DSMs. In Proc. of the 3rd Symp. on Operating Systems Design and Implemen tation (OSDI'99), New Orleans, February 1999. To appear.

....exclusively on independent checkpointing. In particular, we implement independent checkpointing in the context of a software shared memory system to provide a robust, yet fa3 miliar programming model. In the last decade, an impressive amount of research has been conducted in software shared memory [17, 2, 6, 25, 26, 37, 35, 34, 19], mostly aiming for performance (e.g. relaxed consistency models, lazy protocols and communication hardware support) More recently, projects like InterWeave [36] propose a shared memory programming model to support applications that run on wide area clusters of heterogeneous machines. The speci ....

A. Itzkovitz, A. Schuster. MultiView and Millipage - Fine-Grain Sharing in Page-Based DSMs. Proc. 3rd USENIX Symposium on Operating Systems Design and Implementation (OSDI), pp. 215-228, February 1999.

....DOSA for SOR and Water Nsquared on 16 and 32 processors for small and large problem sizes. Figure 9 shows various statistics from the execution of these applications on 32 processors for both problem sizes. 8 Related Work Two other systems have used VM mechanisms for fine grain DSM: Millipede [10] and the Region Trapping Library [4] The fundamental difference between DOSA and these systems is that DOSA takes advantage of a typed 9 language to distinguish a pointer from data at run time and these other systems do not. This allows DOSA to implement a number of optimizations that are not ....

A. Itzkovitz and A. Schuster. Multiview and millipage -- fine-grain sharing in page-based DSMs. In Proceedings of the Third USENIX Symposium on Operating System Design and Implementation, Feb. 1999.

....introduce a protocol (ADSM) that dynamically and automatically associates locks with specific shared pages, in effect achieving part of the behaviour of entry consistency without the need for annotations, but still managing data at page granularity. Brecht and Sandhu [6] and Itzkovitz and Schuster [10] introduce two different techniques for managing data at region granularity without the need for annotations, using a novel pointer swizzling and page fault handling strategy in the former case, and conventional virtual memory fault handling in the latter. Both of these techniques improve the ....

A. Itzkovitz and A. Schuster, "MultiView and Millipage -- Fine-Grain Sharing in Page-Based DSMs", Proceedings of the 3rd Symposium on Operating Systems Design and Implementation (OSDI '99), February, 1999.

....In this work we propose a new approach for on the fly detection of data races in dsm systems. In order to reduce the overhead involved in the detection process, we adapt the detection granularity to that of the dsm sharing unit, and further make use of a method that was proposed recently in [8] to adapt the sharing unit to the application data objects. Indeed, using this approach we were able to drastically reduce the overhead involved in the detection and logging mechanisms. We argue that, in addition to the improved performance, matching the detection and data object granularities can ....

....messages to the execution; the small amount of additional information is piggybacked on the already transferred dsm related messages. We implemented djit using the millipage dsm system, which is capable of sharing in granu2 larity of variables and provides sequentially consistent memory behavior [8]. In order to overcome some of the operating system inefficiencies, we incorporated several optimizations which make novel use of the protection mechanisms. In four of our benchmark applications djit reduces the overhead to only few percent. Another application which was chosen for behaving worst ....

[Article contains additional citation context not shown here]

A. Itzkovitz and A. Schuster. MultiView and Millipage --- Fine-Grain Sharing in PageBased DSMs. In Proc. of the 3rd Symp. on Operating Systems Design and Implementation (OSDI'99), pages 215--228, New Orleans, Feb. 1999.

....the initialization of all mappings. As a practical matter, the three mappings of shared memory region differ in two leading bits of their addresses. Therefore, changing protection is a simple bit masking operation. This approach is superficially similar to the MultiView approach used in Millipede [8], but in fact it is fundamentally different. In MultiView a physical page may be mapped at multiple addresses in the virtual address space, as in SkidMarks, but the similarity ends there. In MultiView, each object resides in its own vpage, which is the size of a VM page. Different 000 000 000 000 ....

....compared our work extensively to TreadMarks [1] and Shasta [11] using them as examples of coarse grained and fine grained DSM systems. Qualitatively similar comparisons can be made with other DSM systems [9, 5, 16] We have also compared our work to the MultiView approach used in Millipede [8]. Dwarkadas et al. 6] compare Cashmere, a coarsegrained system, and Shasta running on an identical platform a cluster of four four way AlphaServers connected by a Memory Channel network. Both systems are designed to leverage the Memory Channel network and take advantage of the hardware shared ....

A. Itzkovitz and A. Schuster. Multiview and millipage -- fine-grain sharing in page-based dsms. In Proceedings of the Third USENIX Symposium on Operating System Design and Implementation, Feb. 1999.

....the additional physical memory overhead incurred by the current RTL implementation. The other major concern is with the RTL interface itself, which requires all pointers to a region to be explicitly declared. In work conducted concurrently and independently of our own, Itzkovitz and Schuster [11] present an alternative approach that provides the functionality that the RTL implements but without having to manipulate pointers to those regions. Itzkovitz and Schuster s MultiView system allocates each region on a separate virtual memory page, but maps each of those regions to the same set of ....

A. Itzkovitz and A. Schuster, "MultiView and Millipage -- Fine-Grain Sharing in Page-Based DSMs", Proceedings of the 3rd Symposium on Operating Systems Design and Implementation (OSDI '99), February, 1999.

....In addition, relaxed consistency protocols introduce substantial additional overhead [17] Fine grain access has been proposed [13, 14] to overcome the false sharing problem, but its implementation usually involves code instrumentation and is limited to certain machine architectures. Recently, in [6], a new technique called MultiView has been proposed to allow fine grain access to shared data without source or binary code modification. This technique makes use of the multiple mapping capabilities of the operating system to provide independent access control to application variables and data ....

....granularity can eliminate false sharing and fragmentation, and reduce unnecessary network traffic. On the other hand, coarse granularity reduces the number of page faults and other expensive consistency protocol operations. Therefore, in addition to setting the granularity level per application [6, 12], performance can be improved by setting different granularity levels in different stages of the execution [11] In this paper we propose a novel technique, called AdaptableView, for providing multiple sharing granularities in page based dsms. AdaptableView extends MultiView by providing ....

[Article contains additional citation context not shown here]

A. Itzkovitz and A. Schuster. MultiView and Millipage --- Fine-Grain Sharing in Page-Based DSMs. In Proc. of the 3rd Symp. on Operating Systems Design and Implementation (OSDI'99), pages 215--228, New Orleans, February 1999.

....Schuster Computer Science Department Technion Israel Institute of Technology assaf cs.technion.ac. il Abstract In this position paper we explore a very recent technique, called MultiView, its applications, and its implications on the design and usage of distributed shared memory systems (dsms) [6]. MultiView can be used to bridge the gap between the large, fixed size memory pages handled by the hardware and operating system, and the relatively small, varying size minipages that are used by applications. Using MultiView, the distributed shared memory system can adapt to the native ....

....concurrent conflicting accesses (which include writes) are expected. Hence, Sequential Consistency (e.g. through the SW MR protocol) is perfectly suited to this situation: it guarantees consistent modifications while avoiding protocolrelated overheads. Indeed, our implementation, as reported in [6], shows initial performance results that support this claim. 3.1 The DSM System Prototype We have implemented a prototype dsm system called millipage. The main goals in building millipage were experimenting with MultiView and exploring the use of a thin protocol layer. The programming model in ....

A. Itzkovitz and A. Schuster. MultiView and Millipage --- Fine-Grain Sharing in Page-Based DSMs. In Proc. of the 3rd Symp. on Operating Systems Design and Implementation (OSDI'99), New Orleans, Feb. 1999. To appear.

No context found.

A. Itzkovitz, A. Schuster, MultiView and Millipage --- fine-grain sharing in page-based DSMs, in: Proceedings of the 1999.

No context found.

A. Itzkovitz and A. Schuster, \MultiView and Millipage { Fine-Grain Sharing in Page-Based DSMs", Proceedings of the 3rd Symposium on Operating Systems Design and Implementation (OSDI '99), February, 1999.

No context found.

Ayal Itzkovitz and Assaf Schuster. Multiview and millipage - fine-grain sharing in page-based DSMs. In Operating Systems Design and Implementation, pages 215--228, 1999.

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