Arvind and R. E. Thomas. "I-Structures: An Efficient Data Type for Functional Languages". Technical Report LCS/TM178, MIT laboratory for Computer Science, 1980.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....which have bounded resource requirements excluding d n to strucre storage, may have unbounded structure storage requirements. The structure storage requirements of a program depend heavily on the model of structures employed, whether it be Dennis s general structures[15,9] or I structures [11] 1. The U interpreter is defined in terms of general structures. The Tagged Token 1General structures require a new mucture, th a single new element, be areted when an element is appended. They are usually implemented as a linked structure, so that pain of the structure can be shared. l structur ....

....limit need not be imposed by CBRs; other resources may be constraining A maximally parallel evaluation requires exponentially more resources than a sequential evaluation. As a more concrete example, consider the matrix multiply program shown in F ure 4 2. It is written in the dataflow language Id [11]. The graph for this pro, ram has three code blocks, one for each of the loops. Collectively these occupy a total of 2.5K bytes of program memory. A single copy of the code can be shared by many invocations. The constant areas are 64, 80, and 112 bytes in length for the outer, middle, and inner ....

[Article contains additional citation context not shown here]

Arvind. and R. E. Thomas. I-Structures: An Efficient Data Type for Functional Languages. 32,4-178, Laboratory for Computer Science, MIT, Cambridge, Mass.,September, 1980.

....1994] In the best case these systems provide the same advantages as data parallel languages. They preserve the abstraction of sequential execution by encapsulating the parallel computation inside routines invoked from a serial program. 7. 3 Functional Languages Functional languages such as Id [Arvind and Thomas 1981] and Sisal [Feo et al. 1990] support a functional model of computation. From the perspective of someone implementing Id or Sisal on a parallel machine, the important feature of these languages is that they allow programs to define the value of a variable at most once. Programs written in Id, ....

Arvind and Thomas, R. 1981. I-structures: An efficient data type for functional languages. Tech. Rep. MIT/LCS/TM-210, Laboratory for Computer Science, Massachusetts Institute of Technology, Cambridge, Mass.

....implementation. The simpler implementation blocks any attempt to do variablevariable binding until at least one of the variables is bound to a value. The price of the simpler implementation is that third party dependencies are not removed in all cases. Futures [Halstead 1985] and I structures [Arvind and Thomas 1980; Veen 1986; Iannucci 1990] resemble this weaker version of logic variables (see Section 9.2.1) There remains a crucial difference with logic variables, namely that Logic Variables for Distributed Computing Delta 9 y f(y1) y f(y1) y f(y1) Centralized Site 1 Site 2 Distributed x y1=foo ....

....along with it. Therefore the above definition is not quite right; to precisely model futures a read only logic variable should be used (see Section 8.1. 3) An I structure (for incomplete structure) is a single assignment array whose elements can be accessed before all the elements are computed [Arvind and Thomas 1980; Veen 1986; Iannucci 1990] It permits concurrency between a computation that calculates the array elements and a computation that uses their values. When the value of an element is needed, then the computation blocks until it is available. An I structure differs from an array of logic variables ....

Arvind and Thomas, R. E. 1980. I-Structures: An efficient data type for functional languages. Tech. Rep. 210, MIT, Laboratory for Computer Science.

....synchronized abstract data type built on top of the version number synchronization mechanism. 1.4 Eliminating mutable data One way to eliminate the synchronization problems associated with explicit concurrency is to eliminate mutable data. Futures (Halstead 1985, Halstead 1986) and I Structures (Arvind and Thomas 1981, Nikhil and Pingali 1989, Nikhil 1990) support implicitly synchronized data structures that are written only once. A future is a place holder that holds the result of a function that may execute concurrently with its caller. Parts of the computation that read the future implicitly suspend until ....

Arvind and Thomas, R. (1981) I-structures: An efficient data type for functional languages. Technical Report MIT/LCS/TM-210, MIT.

....SU synchronizes the threads identified by messages. NIMU manages the routing of messages from nodes, and interfaces to other cluster through the interconnection network. NIMU also manages I structure allocated in SM( Structure Memory) for the global data structure shared among threads. I structure[1] is a special kind of array that same results are returned independent of the order in which reads and writes are performed. I structure was designed to achieve asynchronous parallelism between producer and consumer in handling the large data structure, while preserving determinacy by restricting ....

Arvind and R. E Thomas, "I-structure: An efficient data type for functional languages," Lab. Computer Science MIT Cambridge, Tech. Rep.LSC/TM-178, June 1980.

....is that it contains all context information for that computation. A token contains a continuation and zero or more values. Most network traffic is messages consisting of just tokens. In addition, most transactions across the network are split phase (Tera is a notable exception) I structures [AT80, ANP89] implement non strict arrays. Each memory cell in an I structure memory has an associated presence bit used to tag whether the memory cell contains a valid value or not. This mechanism allows synchronizations to be performed at the array element level, rather than at the array level, by ....

....In [NRB93] the basic block algorithm that was described did not make branches terminal which required additional check for deadlock free executions. 3 In a non strict structure store access, an element of a structure may be read before the entire structure is written, and requires I structure [AT80] type of support. 27 BRR ADR ADR a b MLR c d 1 2 3 4 2.0 (constant) Figure 3.1: If then else merge 2. Deadlock cases are resolved during the fission stage by checking for dynamic and static cycles in each cluster. Static cycles are detected by starting from each result of each cluster and ....

Arvind and R. E. Thomas. I-structures: An efficient data type for functional languages. Technical Report LCS/TM-178, MIT, Laboratory for Computer Science, 1980.

....is needed, has proven to be the Achilles heel of the implementations. 2.3.1 Incremental Data Structures Let us consider how the dynamic dataflow model deals with manipulating complex data structures. The most common approach is to use what are known as I structures (for incremental structures) [7]. The basic idea is to associate with each atomic unit of a data structure with status bits and a queue of deferred reads. The status of the atomic unit can be one of PRESENT, which means that it can be read but not written, ABSENT, which means that read has to be deferred but it can be written, ....

Arvind and R.E. Thomas. I-structures: An efficient data type for functional languages. Technical Report MIT/LCS/TM-210, Laboratory of Computer Science, M.I.T., October 1981.

....RISC processors. Data flow research continues with Monsoon [9] and other modern data flow machines. While data flow machines handled the transfer of data tokens well, they had trouble with large arrays, because it was not reasonable to send copies of such arrays around the system. I structures [10] were added to solve the problem. I structures provide storage space on certain dedicated nodes in the system to hold arrays. I structure fetch requests are split phase operations consisting of two messages: a request to the I structure storage node, and a response with the data. If the ....

Arvind and R. E. Thomas, I-Structures: An Efficient Data Type For Functional Languages, Technical Report LCS/TM-178, MIT Laboratory for Computer Science, 17

....Then by a transitive closure algorithm, the dependence sets are built. 3. All nodes with the same dependence sets are then collected into one cluster. 3 In a non strict structure store access, an element of a structure may be read before the entire structure is written, and requires I structure [3] type of support. CLUSTER A ADR SYN CLUSTER C CLUSTER B 1 CLI EXR 1A WSS SYN ADI RSS b dynamic cycle a static cycle Figure 2: Deadlock potential due to circular dependency 4.3 Optimization (Fusion Phase) Both the basic block and dependence sets algorithms are followed by an optimization ....

Arvind and R. E. Thomas. I-structures: An efficient data type for functional languages. Technical Report LCS/TM-178, MIT, Laboratory for Computer Science, 1980.

....but can be applied to MIMD machines as well. Programs running on SPMD machines typically need access to large arrays, lists, or trees of data, often larger than the memory available in a single node. These data structures may either be stored on special purpose data storage nodes (I Structure [6] nodes, for example) or partitioned and distributed across the system. When a program running on a given node needs to access data that is not in the node s local memory, it performs a remote fetch operation to get the data. A fetch consists of a request message to the node where the data resides ....

....Data flow research continues with Monsoon [8] Sigma 1 [9] and other recent data flow machines. While data flow machines handled the transfer of data tokens well, they had trouble with large arrays, because it is not reasonable to send copies of arrays around the system. I structures [6] were added to solve the problem. They provide storage on dedicated nodes to hold arrays of data items. I structure fetch requests are split phase operations consisting of a request message to an I structure node and a response message with the data. If the requested data item is not yet ....

Arvind and R. E. Thomas, "I--Structures: An efficient data type for functional languages," Technical Report LCS/TM--178, MIT Laboratory for Computer Science, 1980.

....until the element is defined by a write operation. We will not dwell here on the details of I structures. Suffice it to say that I structure memory also supports conventional reads and writes of global memory locations using the split phase protocol explained below. For more on I structures, see [14, 13]. As shown in Figure 11, a Monsoon computer consists of a mixture of Processor and IStructure Units, interconnected by a packet switched network. The processor units include memory for instructions and activation frames; the I structure units contain the memory for large arrays and other data ....

Arvind and Robert E. Thomas, "I-structures: An efficient data type for functional languages, " Technical Memo MIT/LCS/TM-178, MIT Laboratory for Computer Science, September 1980.

No context found.

Arvind and R. E. Thomas. "I-Structures: An Efficient Data Type for Functional Languages". Technical Report LCS/TM178, MIT laboratory for Computer Science, 1980.

No context found.

Arvind, and R. E. Thomas. I-Structures: An Efficient Data Type for Functional Languages. Tech. Rep. TM-178, Laboratory for Computer Science, MIT, Cambridge, Mass., September, 1980.

No context found.

Arvind, R. E. Thomas, "I-Structures: An efficient Data Type for Functional Languages ", TM-178, Laboratory for Computer Science, MIT, Cambridge, 1980

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