Active Messages Research Articles

A Communication Library Using Active Messages to Improve Performance of PVM

We present a communication library to improve performance of PVM. The new library introduces communication primitives based on Active Messages. We propose a hybrid scheme that includes a signal driven message notification scheme plus controlled polling. The new communication library is tested along with the normal PVM library to assess the improvement in performance.

Security considerations for active messages

Messages are generally regarded as passive since they merely act as a conveyance for information. An active message triggers processing activities at each of its recipients. A discussion is provided on the different classes of processing activities that can occur for active messages. A mechanism for ensuring the security of an active message and its associated processing operations is also outlined.

A comparison of different message-passing paradigms for the parallelization of two irregular applications

We present experimental results for parallelizing two breadth-first search-based applications on Thinking Machines CM-5 by using two different message-passing paradigms, one based onsend/receive and the other based onactive messages. The parallelization of these applications requires fine-grained communication. Our results show that the active messages-based implementation gives significant improvement over the send/receive-based implementation. The improvements can largely be attributed to the lower latency of the active messages implementation.

Assessing fast network interfaces

Assessing the performance of emerging high-speed networks is difficult. Our communication microbenchmark evaluates design changes in a communications system. Our microbenchmark generates a graphical signature from which we extract communication performance parameters for the hardware-software tandem. We use the LogP conceptual model for communication systems. (LogP stands for the parameters latency, overhead, gap, and processors). We study three important platforms that represent diverse points in the network interface design space: the Intel Paragon, Meiko CS-2, and a cluster of Sparcstation-20s connected by Myrinet switches using LANai SBus cards. Our study views each machine as a gray box supporting Active Messages and conforming to the LogP framework. We devise a simple set of communication microbenchmarks and measure the performance on each platform to obtain the LogP parameters.

Optimistic active messages

Low-overhead message passing is critical to the performance of many applications. Active Messages reduce the software overhead for message handling: messages are run as handlers instead of as threads, which avoids the overhead of thread management and the unnecessary data copying of other communication models. Scheduling the execution of Active Messages is typically done by disabling and enabling interrupts, or by polling the network. This primitive scheduling control, combined with the fact that handlers are not schedulable entities, puts severe restrictions on the code that can be run in a message handler. This paper describes a new software mechanism, Optimistic Active Messages (OAM), that eliminates these restrictions; OAMs allow arbitrary user code to execute in handlers, and also allow handlers to block. Despite this gain in expressiveness, OAMs perform as well as Active Messages. We used OAM as the base for an RPC system, Optimistic RPC (ORPC), for the Thinking Machines CM-5 multiprocessor; it consists of an optimized thread package and a stub compiler that hides communication details from the programmer. ORPC is 1.5 to 5 times faster than traditional RPC (TRPC) for small messages and performs as well as Active Messages (AM). Applications that primarily communicate using large data transfers or are fairly coarse-grained perform equally well, independent of whether AMs, ORPCs, or TRPCs are used. For applications that send many short messages, however, the ORPC and AM implementations are up to three times faster than the TRPC implementations. Using ORPC, programmers obtain the benefits of well-proven programming abstractions such as threads, mutexes, and condition variables, do not have to be concerned with communication details, and yet obtain nearly the performance of hand-coded Active Message programs.

Evaluating the locality benefits of active messages

A major challenge in fine-grained computing is achieving locality without excessive scheduling overhead. We built two J-Machine implementations of a fine-grained programming model, the Berkeley Threaded Abstract Machine. One implementation takes an Active Messages approach, maintaining a scheduling hierarchy in software in order to improve data cache performance. Another approach relies on the J-Machine's message queues and fast task switch, lowering the control costs at the expense of data locality. Our analysis measures the costs and benefits of each approach, for a variety of programs and cache configurations. The Active Messages implementation is strongest when miss penalties are high and for the finest-grained programs. The hardware-buffered implementation is strongest in direct-mapped caches, where it achieves substantially better instruction cache performance.

Export of Abscisic Acid, 1-Aminocyclopropane-1-Carboxylic Acid, Phosphate, and Nitrate from Roots to Shoots of Flooded Tomato Plants (Accounting for Effects of Xylem Sap Flow Rate on Concentration and Delivery).

We determined whether root stress alters the output of physiologically active messages passing from roots to shoots in the transpiration stream. Concentrations were not good measures of output. This was because changes in volume flow of xylem sap caused either by sampling procedures or by effects of root stress on rates of whole-plant transpiration modified concentrations simply by dilution. Thus, delivery rate (concentration x sap flow rate) was preferred to concentration as a measure of solute output from roots. To demonstrate these points, 1-aminocyclopropane-1-carboxylic acid (ACC), abscisic acid, phosphate, nitrate, and pH were measured in xylem sap of flooded and well-drained tomato (Lycopersicon esculentum Mill., cv Ailsa Craig) plants expressed at various rates from pressurized detopped roots. Concentrations decreased as sap flow rates were increased. However, dilution of solutes was often less than proportional to flow, especially in flooded plants. Thus, sap flowing through detopped roots at whole-plant transpiration rates was used to estimate solute delivery rates in intact plants. On this basis, delivery of ACC from roots to shoots was 3.1-fold greater in plants flooded for 24 h than in well-drained plants, and delivery of phosphate was 2.3-fold greater. Delivery rates of abscisic acid and nitrate in flooded plants were only 11 and 7%, respectively, of those in well-drained plants.

Low-latency communication over ATM networks using active messages

Today's communication architectures for parallel machines reduce communication overheads and latencies by over an order of magnitude. However, carrying over these techniques to workstation clusters connected by an ATM network presents major design challenges. We discuss the differences in communication characteristics between workstation clusters built from standard hardware and software components and state-of-the-art multiprocessors, and then evaluate a prototype implementation of an active message communication layer. Application round-trip latencies of about 50 microseconds for small messages roughly compare to a similar implementation on the Thinking Machines CM-5 multiprocessor.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Software overhead in messaging layers

Despite improvements in network interfaces and software messaging layers, software communication overhead still dominates the hardware routing cost in most systems. In this study, we identify the sources of this overhead by analyzing software costs of typical communication protocols built atop the active messages layer on the CM-5. We show that up to 50–70% of the software messaging costs are a direct consequence of the gap between specific network features such as arbitrary delivery order, finite buffering, and limited fault-handling, and the user communication requirements of in-order delivery, end-to-end flow control, and reliable transmission. However, virtually all of these costs can be eliminated if routing networks provide higher-level services such as in-order delivery, end-to-end flow control, and packet-level fault-tolerance. We conclude that significant cost reductions require changing the constraints on messaging layers: we propose designing networks and network interfaces which simplify or replace software for implementing user communication requirements.

Software overhead in messaging layers

Despite improvements in network interfaces and software messaging layers, software communication overhead still dominates the hardware routing cost in most systems. In this study, we identify the sources of this overhead by analyzing software costs of typical communication protocols built atop the active messages layer on the CM-5. We show that up to 50–70% of the software messaging costs are a direct consequence of the gap between specific network features such as arbitrary delivery order, finite buffering, and limited fault-handling, and the user communication requirements of in-order delivery, end-to-end flow control, and reliable transmission. However, virtually all of these costs can be eliminated if routing networks provide higher-level services such as in-order delivery, end-to-end flow control, and packet-level fault-tolerance. We conclude that significant cost reductions require changing the constraints on messaging layers: we propose designing networks and network interfaces which simplify or replace software for implementing user communication requirements.

Solving microstructure electrostatistics with MIMD parallel supercomputers and Split-C

We consider parallel computational algorithms for the boundary integral solutions of the Laplace equation for use in the simulation of electrorheological fluids and as a model study of a class of elliptic partial differential equations that appear in basic microscopic descriptions of heterogeneous structured continua. The viewpoint is that of constructing large scale simulations that bridge micro- and macro-length and time scales on state-of-the-art parallel supercomputers. Because of long range interactions, fast communications are the key to scalable N-Body algorithms. The communication scheduling strategies of Fuentes and Kim are examined in two contexts on the Thinking Machines CM-5 parallel computer. First, an N-Body simulation implementation in C using the standard send-receive message passing primitives in the CMMD 2.0 library shows that communication scheduling leads to dramatic improvements in computational performance. Second, an implementation in Split-C, which uses highly efficient active messages to implement shared memory communication, reduces communication overhead by an order of magnitude. Taken together, these two developments portend great promise for the development of efficient large scale simulations using portable, high level parallel programming languages.

CMMD: Active messages on the CM-5

Active messages provide an important new communication primitive for building message passing systems. CMMD, the message passing system of the CM-5, uses active messages as a basic subtrate for constructing multiple, low overhead, communication paradigms. Examples are given which show how developers may incorporate active messages in application-specific ways.

In situ localization of PCR-amplified human and viral cDNAs.

We describe a technique, called reverse transcriptase (RT) in situ PCR, whereby RNA may be nonisotopically detected in fixed cells when amplified by PCR after cDNA synthesis by RT. RT in situ PCR using primers specific for the measles virus generated an intense signal in most measles-infected HeLa cells, as compared to the weak signal generated in few cells using standard in situ hybridization analysis. The viral RNA that localized to the nucleus spared the nucleoli, was most evident when the RT step used the primer complementary to the negative genomic strand, and was demonstrated in all multinucleated cells and the majority of uninucleate cells. A hybridization signal was evident with standard RNA in situ hybridization using the human megakaryocyte cell line Dami and a probe for glycoprotein IIB (GIIB) mRNA but not a probe for amyloid precursor protein (APP) or gelsolin (GEL) mRNA. After RT in situ PCR, signals were evident for each target localizing to the nucleolus for APP and to perinucleolar and cytoplasmic locations for GEL and GIIB. The latter findings suggest that mRNAs may follow different geographic pathways as they progress from premessage to transcriptionally active message.

An Intelligent Economic Decision Support System IEDSS for Development Strategic Design

Abstract The paper presents the theoretical background and the conceptual framework for designing an intelligent decision support system towards the economic development strategy of China (national and/or regional). AI technology plays prominent and direct roles in the implementation of the IEDSS. The IEDSS performs its solution searching process as learning. When conducting a problem solving, the decision maker DM and the system DSS compensate each other with their equipped knowledge capability and work cooperatively to reach an optimal or sub-optimal solution. The “intelligent” features of the system are configured with an integrated knowledge base management system IKBMS, and an intuitive man-machine interface M-M. The IKBMS is the machine brain of the artificial system, and knowledge is the brain’s internal representation of the external world. And the system’s model-based aspiration led problem solving function is just its knowledgeable behavior. The M-M monitors all user actions during the problem solving process, and offers user—requested passive help, program-requested active prompting and error— driven active messaging in a uniform fashion. The IKBMS is essentially an object- oriented (or frame / rule—based) system and the desired M-M functions are dynamicly determined by the procedure embedded in the object frame slots. The process of developing a solution in the managerial domain is mostly evolutionary or incremental rather than “one-shot”. The reported M-M of IEDSS is designed to facilitate easy utilization and access of the system in a way that is object-oriented.

Active messages

The design challenge for large-scale multiprocessors is (1) to minimize communication overhead, (2) allow communication to overlap computation, and (3) coordinate the two without sacrificing processor cost/performance. We show that existing message passing multiprocessors have unnecessarily high communication costs. Research prototypes of message driven machines demonstrate low communication overhead, but poor processor cost/performance. We introduce a simple communication mechanism, Active Messages , show that it is intrinsic to both architectures, allows cost effective use of the hardware, and offers tremendous flexibility. Implementations on nCUBE/2 and CM-5 are described and evaluated using a split-phase shared-memory extension to C, Split -C. We further show that active messages are sufficient to implement the dynamically scheduled languages for which message driven machines were designed. With this mechanism, latency tolerance becomes a programming/compiling concern. Hardware support for active messages is desirable and we outline a range of enhancements to mainstream processors.

Virtual-channel flow control

Network throughput can be increased by dividing the buffer storage associated with each network channel into several virtual channels. Each physical channel is associated with several small queues, virtual channels, rather than a single deep queue. The virtual channels associated with one physical channel are allocated independently but compete with each other for physical bandwidth. Virtual channels decouple buffer resources from transmission resources. This decoupling allows active messages to pass blocked messages using network bandwidth that would otherwise be left idle. The paper studies the performance of networks using virtual channels using both analysis and simulation. These studies show that virtual channels increase network throughput, by a factor of four for 10-stage networks, and reduce the dependence of throughput on the depth of the network. >

The Coordination of Distributed Active Messages in a Dynamic Network Topology

An active message is a message that can interact with the recipient, as opposed to a 'conventional' message, which only carries data from one place to another. As a result of the response obtained from the recipient, an active message can decide to send copies of itself to other destinations. The coordination of these copies is a problem, especially in large networks, because the copies can be sent to destinations anywhere in a network. In this article we present a model that establishes coordination between the distributed copies of an active message in a dynamic network topology, dynamic in the sense that nodes may go down or new nodes can be added to the network. We elaborate on an earlier version of the model, which was only applicable to networks with a static topology.

Active transcription of the selenium-dependent glutathione peroxidase gene in selenium-deficient rats

Selenium-dependent glutathione peroxidase (Se-GSH-Px, Ec.1.11.1.9) is the best characterized selenoenzyme in higher animals. However, neither the mechanism whereby selenium (Se) becomes incorporated into the enzyme nor the level at which the expression of Se-GSH-Px gene is regulated by Se is fully understood. In the current investigation, we have determined the relative rates of the transcription of the Se-GSH-Px gene in purified liver nuclei isolated from rats fed on Se-supplemented or Se-deficient diets. No significant difference in the transcription rates appeared in these two groups. These results are consistent with the previous observations that active message for Se-GSH-Px — that is, translatable mRNA for Se-GSH-Px — is present in Se-deficient tissues (Li et al., J. Biol. Chem., 265, 108–113, 1990). The data also suggest that the alteration of Se-GSH-Px activity and the corresponding protein and mRNA levels in rats subjected to dietary Se manipulation can be attributed only to post-transcriptional regulation.

Intelligent Help for CACE Applications

An "intelligent", user adaptive and context-sensitive help system for Computer Aided Control Engineering (CACE) applications is presented, The developed system attempts to address the need for better structured and more active help systems to cope with the growing operational and functional complexIty of modern CACE programs. It monitors all user actions to determine the knowledge level of the user within each specific domain and adjusts its own behaviour accordmgly. User-requested passive help, program-requested active prompting and error-driven active messaging are all supported in a uniform fashion.The system consists of a generic kernel driving the help system in a knowledge-domain independent fashion, and a domain-specific knowledge base. A prototype of the system has been implemented in Prolog.

Virtual-channel flow control

Network throughput can be increased by dividing the buffer storage associated with each network channel into several virtual channels [DalSei]. Each physical channel is associated with several small queues, virtual channels, rather than a single deep queue. The virtual channels associated with one physical channel are allocated independently but compete with each other for physical bandwidth. Virtual channels decouple buffer resources from transmission resources. This decoupling allows active messages to pass blocked messages using network bandwidth that would otherwise be left idle. Simulation studies show that, given a fixed amount of buffer storage per link, virtual-channel flow control increases throughput by a factor of 3.5, approaching the capacity of the network.