On profitability and maximum tolerable latency in the high-frequency trading of a microtrend anomaly

Multithreaded multiprocessor systems (MMS) have been proposed to tolerate long latencies for communication. This paper provides an analytical framework based on closed queueing networks to quantify and analyze the latency tolerance of multithreaded systems. We introduce a new metric, called the tolerance index, which quantifies the closeness of performance of the system to that of an ideal system. We characterize the latency tolerance with the changes in the architectural and program workload parameters. We show how an analysis of the latency tolerance provides an insight to the performance optimizations of fine grain parallel program workloads.

A person often uses a single search engine for very different tasks. For example, an author editing a manuscript may use the same academic search engine to find the latest work on a particular topic or to find the correct citation for a familiar article. The author's tolerance for latency and accuracy may vary according to task. However, search engines typically employ a consistent approach for processing all queries. In this paper we explore how a range of search needs and expectations can be supported within a single search system using differential search. We introduce CiteSight, a system that provides personalized citation recommendations to author groups that vary based on task. CiteSight presents cached recommendations instantaneously for online tasks (e.g., active paper writing), and refines these recommendations in the background for offline tasks (e.g., future literature review). We develop an active cache-warming process to enhance the system as the author works, and context-coupling, a technique for augment sparse citation networks. By evaluating the quality of the recommendations and collecting user feedback, we show that differential search can provide a high level of accuracy for different tasks on different time scales. We believe that differential search can be used in many situations where the user's tolerance for latency and desired response vary dramatically based on use.

Rather than improving single-threaded performance, with the dawn of the multi-core era, processor microarchitects have exploited Moore's law transistor scaling by increasing core density on a chip and increasing the number of thread contexts within a core. However, single-thread performance and efficiency is still very relevant in the power-constrained multi-core era, as increasing core counts do not yield corresponding performance improvements under real thermal and thread-level constraints. This dissertation provides a detailed study of register reference count structures and its application to both conventional and non-conventional, latency tolerant, out-of-order processors. Prior work has incorporated reference counting, but without a detailed implementation or energy model. This dissertation presents a working implementation of reference count structures and shows the overheads are low and can be recouped by the techniques enabled in high-performance out-of-order processors. A study of register allocation algorithms exploits register file occupancy to reduce power consumption by dynamically resizing the register file, which is especially important in the face of wider multi-threaded processors who require larger register files. Latency tolerance has been introduced as a technique to improve single threaded performance by removing cache-miss dependent instructions from the execution pipeline until the miss returns. This dissertation introduces a microarchitecture with a predictive approach to identify long-latency loads, and reduce the energy cost and overhead of scaling the instruction window inherent in latency tolerant microarchitectures. The key features include a front-end predictive slice-out mechanism and in-order queue structure along with mechanisms to reduce the energy cost and register-file usage of executing instructions. Cycle-level simulation shows improved performance and reduced energy delay for memory-bound workloads. Both techniques scale processor resources, addressing register file inefficiency and the allocation of processor resources to instructions during low ILP regions.

Why are shared memory MIMD computers with many processors difficult to implement? The answer depends in part on the definition of the term “shared memory”, but there is probably broad agreement that any sort of shared memory system with hundreds of processors is a challenge. The main problem is the latency associated with memory access and its consequences for processor performance. There are two possible solutions to this problem, namely latency avoidance and latency tolerance. Latency avoidance is accomplished by arranging a processor’s memory accesses so that most of them are to locations that are both spatially and temporally nearby. Latency tolerance is brought about through the use of additional parallelism. Both ideas have been used in shared memory systems, with varying success.

This paper provides quantitative measurements of load latency tolerance in a dynamically scheduled processor. To determine the latency tolerance of each memory load operation, our simulations use flexible load completion policies instead of a fixed memory hierarchy that dictates the latency. Although our policies delay load completion as long as possible, they produce performance (instructions committed per cycle (IPC)) comparable to an ideal memory system where all loads complete in one cycle. Our measurements reveal that to produce IPC values within 8 % of the ideal memory system, between 1 % and 62 % of loads need to be satisfied within a single cycle and that up to 84 % can be satisfied in as many as 32 cycles, depending on the benchmark and processor configuration. Load latency tolerance is largely determined by whether an unpredictable branch is in the load’s data dependence graph and the depth of the dependence graph. Our results also show that up to 36 % of all loads miss in the level one cache yet have latency demands lower than second level cache access times. We also show that up to 37 % of loads hit in the level one cache even though they possess enough latency tolerance to be satisfied by lower levels of the memory hierarchy. 1

Latency tolerance is one of main problems of software synthesis in the design of hardware-software mixed systems. This paper presents a methodology for speeding up systems through latency tolerance which is obtained by decomposition of tasks and generation of an efficient scheduler. The task decomposition process focuses on the dependency analysis of system i/o operations. Scheduling of the decomposed tasks is performed in a mixed static and dynamic fashion. Experimental results show the significance of our approach.

Wireless sensor networks (WSNs) developed for the monitoring of critical military or civilian infrastructures are expected to have long life cycle with ultra-low power consumption. An ultra-low power wireless sensing scheme is developed by exploiting the unique features of infrastructure monitoring systems, which usually have long latency tolerance, low data rate, and strong correlation among data collected by spatially distributed sensors. The wireless sensor nodes asynchronously transmit measured data through a new exponential-interval media access control (EI-MAC) scheme, which can asymptotically almost surely (a.a.s.) achieve collision-free communication by leveraging on the long latency tolerance and low data rate of the system. Two low power sensing schemes, namely, compressive detection (CD) and compressive transmission (CT), are proposed in recognition of the strong correlation among data samples collected by n spatially distributed sensing nodes. Both the two schemes are fully scalable; have ultra-low power consumption; have less distortion compared to conventional schemes; and allow the sensing nodes to operate asynchronously without central control. Theoretical analysis shows that the normalized mean square distortion of the recovered information scales as.

Latency tolerance is one of main problems of software synthesis in the design of hardware-software mixed systems. This paper presents a methodology for speeding up systems through latency tolerance which is obtained by decomposition of tasks and generation of an efficient scheduler. The task decomposition process focuses on the dependency analysis of system i/o operations. Scheduling of the decomposed tasks is performed in a mixed static and dynamic fashion. Experimental results show the significance of our approach.

In both hardware-only and software-only directory protocols the performance is often limited by memory access stall times. To increase the performance, several latency tolerating and reducing techniques have been proposed and shown effective for hardware-only directory protocols. For software-only directory protocols, the efficiency of a technique depends not only on how effective it is as seen by the local processor but also on how it impacts the software handler execution overhead in the node where a memory block is allocated. Based on architectural simulations and case studies of three techniques, we find that prefetching can degrade the performance of software-only directory protocols due to useless prefetches. A relaxed memory consistency model hides all write latency for software-only directory protocols, but the software handler overhead is virtually unaffected and now constitutes a larger portion of the execution time. Overall, latency tolerating techniques for software-only directory protocols must be chosen with more care than for hardware-only directory protocols.

This paper provides quantitative measurements of load latency tolerance in a dynamically scheduled processor. To determine the latency tolerance of each memory load operation, our simulations use flexible load completion policies instead of a fixed memory hierarchy that dictates the latency. Although our policies delay load completion as long as possible, they produce performance (instructions committed per cycle (IPC)) comparable to an ideal memory system where all loads complete in one cycle. Our measurements reveal that to produce IPC values within 8% of the ideal memory system, between 1% and 62% of loads need to be satisfied within a single cycle and that up to 84% can be satisfied in as many as 32 cycles, depending on the benchmark and processor configuration. Load latency tolerance is largely determined by whether an unpredictable branch is in the load's data dependence graph and the depth of the dependence graph. Our results also show that up to 36% of all loads miss in the level one cache yet have latency demands lower than second level cache access times. We also show that up to 37% of loads hit in the level one cache even though they possess enough latency tolerance to be satisfied by lower levels of the memory hierarchy.

High performance superscalar architectures used to exploit instruction level parallelism in single-thread applications have become too complex and too power hungry for the many-core processors era. We propose a new architecture that uses multiple latency-tolerant in-order cores to improve single-thread performance, without requiring complex out-of-order execution hardware or large, power hungry register files and instruction buffers. Using simple cores to provide improved single-thread performance for conventional difficult-to-parallelize applications allows designers to place many more of these cores on the same die. Consequently, emerging highly parallel applications can take full advantage of the many-core parallel hardware without sacrificing performance of inherently serial applications.Our architecture splits single-thread program execution into disjoint control and data threads that execute concurrently on multiple latency-tolerant in-order cores. Hence we call this style of execution Disjoint Out-of-Order Execution (DOE). DOE is a novel implementation of Speculative Multithreading (SpMT). It uses latency tolerance to overcome performance issues of SpMT caused by load imbalance and inter-thread data communication delays.Using control independence prediction hardware to spawn threads, we simulate the potential performance of DOE on a subset of Spec2000 integer benchmarks under various parallelism scenarios and for DOE configurations of 2, 4, 6 and 8 single-issue latency tolerant cores.

Latency tolerance is one of the main concerns in parallel processing. Data Driven Multithreading, a technique that uses extra hardware to schedule threads for execution based on data availability, allows for better performance, through latency tolerance. With Data Driven Multithreading a thread is scheduled for execution only if all of its inputs have been produced and placed in the processor’s local memory. Communication and synchronization are decoupled from the computation portions of a program, i.e. they execute asynchronously. Thus, no synchronization or communication latencies will be experienced. The processor can, though be idle when there are no threads ready for execution, Thus, communication latencies are difficult to hide completely in applications with high communication-to-computation characteristics.This paper presents three mechanisms for the implementation of the communication assist of a Data Driven Multithreaded architecture. The first mechanism relies only on fine grain communication, where each packet can transfer a single value. With the second mechanism, the communication assist is modified to support block data communication through the same fine grain interconnection network of the first configuration. The third mechanism employs a broadcast network such as Ethernet to transfer blocks of data, while fine grain communication is handled the same way as with the other two mechanisms.KeywordsInterconnection NetworkProcessing NodeCommunication LatencyVirtual CircuitEthernet NetworkThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Latency in virtual reality (VR) can decrease the feeling of presence and body ownership. How users perceive latency, however, is malleable and affected by the design of the virtual content. Previous work found that an avatar’s visual appearance, particularly its perceived fitness, can be leveraged to change user perception and behavior. Moreover, previous work investigating non-VR video games also demonstrated that controlling avatars that visually conform to users’ expectations associated with the avatars’ perceived characteristics increases the users’ latency tolerance. However, it is currently unknown if the avatar’s visual appearance can be used to modulate the users’ latency sensitivity in full-body motion-tracked VR. Therefore, we conducted two studies to investigate if the avatars’ appearance can be used to decrease the negative impact of latency. In the first study, 41 participants systematically determined two sets of avatars whose visual appearance is perceived to be more or less fit in two physically challenging tasks. In a second study (N = 16), we tested the two previously determined avatars (perceived to be more fit vs. perceived to be less fit) in the two tasks using VR with two levels of controlled latency (system vs. high). We found that embodying an avatar perceived as more fit significantly increases the participants’ physical performance, body ownership, presence, and intrinsic motivation. While we show that latency negatively affects performance, our results also suggest that the avatar’s visual appearance does not alter the effects of latency in VR.

Advances in the computational sciences and AI have been critical in simulating the projected impacts of climate change in agriculture together with quantifying mitigation strategies to reduce agriculture’s influence on climate change. The role of computation in bringing understanding to agroecosystem change from the farm to national scales is pervasive, being central for remote and ground IoT sensing, big data analysis, process simulation, data assimilation, the capture of error and uncertainty, sensor network design and interactive visualization of high dimensional outputs. Component farm processes for soils, plants, livestock, biodiversity, and water and gaseous emissions are multi-scale and casual with complex space-time connections. Respective component datasets, measured or sensed optimally and adaptively in space and time, are needed to dynamically inform model simulations for past, present, and future states. This can be couched within a scale-aware, farm decision support tool (DSTs - say, via a digital model, shadow, or twin), where its virtual-world forecasts can inform both on-farm decisions and, via extension to networked farm DSTs, farming policy. Given an accurate capture of uncertainty, the risk associated with farm management decisions can be quantified facilitating planning for sustainable farming in the long-term, coupled with (short-term) early warning signals for diminishing system resilience to increasing threats of abiotic, biotic, and other stresses. However, while such AI-driven technologies in agriculture often appear good on paper, what is actually viable in practice? (Rose et al. 2016) noted the existence of 395 agricultural DSTs and recommended 15 of them for their effective design and delivery. Suffice to say, not all of the 395 DSTs are still available, nine years on. Further, the reviewed DSTs were not all sofware-based (on- or off-line) but included those that were paper-based. A question then arises, does AI implicitly change what is possible, practical, and useful for land managers, farmers and their advisors? Or do the same inherent limitations of DSTs remain? In turn, how does this translate to effective government agricultural policy? This paper seeks to elucidate on such questions through a consideration of the following: What advantages might an AI-driven DST (AI-DST) have when compared to one using older technologies? What data does the farmer need to collect and manage to support the AI-DST? What are the minimum data requirements and at what cost? Does the AI-DST provide functionality for cost benefit analyses for the on-going relevance of the data collected? How do the AI-DST outputs reflect data quality, error, and sparseness? Does the scale of on-farm measurements match the scale of the sampled process and subsequent scales of decision making? Is data capture timely enough with tolerable latency? For an AI-DST scenario evaluation – are the interplays between different management and climate scenarios fully described and caveated for practical use? How can the farmer be empowered with their intrinsic expert knowledge of their farm or their farming philosophy within the AI-DST? Is an AI-DST only ever complementary? What about an AI-DST with human in the loop AI? What training and support are required for AI-DST use? How does the farmer know if decisions informed by the AI-DST are beneficial - especially in the long run? Has the weather just been coincidently beneficial? Does the AI-DST capture the interplay between farm profitability and government payments or incentives. How does this interplay vary over time, different geographies, and for different farm practices? How should a farmer proceed with conflicting advice when using multiple AI-DSTs with different objectives. For example, priority decisions from an AI-DST for soil health may conflict with those from an AI-DST for field margin biodiversity. Is there a ‘one-size-fits’ all AI-DST? Is this AI-DST desirable? What are the options for retraining / revising / updating a given AI-DST’s model framework and software given ever changing challenges of climate change; for example, are extremes (drought, floods) or extreme changes (i.e. winter one day, summer the next) in the weather the greater problem? Is the AI-DST typically on board with or resilient to the Zeitgeist (e.g., a 'world without livestock' debate). Does the AI-DST allow for alignment with digital innovations, in say soil sensing, robotics? Does the AI-DST adequately capture and explain concepts of decision risk? Are the AI-DST's output (and input) visualisations relevant and appropriate, in this respect? Given not all farms, farmers and their advisors are made equal – how does the AI-DST cater for this? How does the AI-DST capture the inherently diverse nature of farming? Are AI-DST informed decisions to be made by a farm owner, farm tenant or farm manager? Does the AI-DST cater for a given farm’s route to market and when and where are these routes optimal? How does the AI-DST capture, and adapt to, unintended consequences of the decisions made? Similarly, how does the AI-DST respond to unforeseen external influences, such as international conflict, widespead drought? Are AI-DST-based decisions effective across multiple scales – benefitting individual farms and networked farms alike? Does the AI-DST promote use within farmer networks, community of practise and farmer cooperation? For example, AI-DSTs informed by shared data resources amongst farms within the same catchment. How do on-farm decisions influence the food supply chain – from farm to fork? Does the AI-DST provide this broader picture functionality? For example, does the AI-DST provide options, not only for farm productivity and farm emissions but also those concerned with externalities such as energy use, transportational costs, societal effects and more? What can be learnt and transferred from AI-DSTs and non-AI DSTs in other domains? What advantages might an AI-driven DST (AI-DST) have when compared to one using older technologies? What data does the farmer need to collect and manage to support the AI-DST? What are the minimum data requirements and at what cost? Does the AI-DST provide functionality for cost benefit analyses for the on-going relevance of the data collected? How do the AI-DST outputs reflect data quality, error, and sparseness? Does the scale of on-farm measurements match the scale of the sampled process and subsequent scales of decision making? Is data capture timely enough with tolerable latency? For an AI-DST scenario evaluation – are the interplays between different management and climate scenarios fully described and caveated for practical use? How can the farmer be empowered with their intrinsic expert knowledge of their farm or their farming philosophy within the AI-DST? Is an AI-DST only ever complementary? What about an AI-DST with human in the loop AI? What training and support are required for AI-DST use? How does the farmer know if decisions informed by the AI-DST are beneficial - especially in the long run? Has the weather just been coincidently beneficial? Does the AI-DST capture the interplay between farm profitability and government payments or incentives. How does this interplay vary over time, different geographies, and for different farm practices? How should a farmer proceed with conflicting advice when using multiple AI-DSTs with different objectives. For example, priority decisions from an AI-DST for soil health may conflict with those from an AI-DST for field margin biodiversity. Is there a ‘one-size-fits’ all AI-DST? Is this AI-DST desirable? What are the options for retraining / revising / updating a given AI-DST’s model framework and software given ever changing challenges of climate change; for example, are extremes (drought, floods) or extreme changes (i.e. winter one day, summer the next) in the weather the greater problem? Is the AI-DST typically on board with or resilient to the Zeitgeist (e.g., a 'world without livestock' debate). Does the AI-DST allow for alignment with digital innovations, in say soil sensing, robotics? Does the AI-DST adequately capture and explain concepts of decision risk? Are the AI-DST's output (and input) visualisations relevant and appropriate, in this respect? Given not all farms, farmers and their advisors are made equal – how does the AI-DST cater for this? How does the AI-DST capture the inherently diverse nature of farming? Are AI-DST informed decisions to be made by a farm owner, farm tenant or farm manager? Does the AI-DST cater for a given farm’s route to market and when and where are these routes optimal? How does the AI-DST capture, and adapt to, unintended consequences of the decisions made? Similarly, how does the AI-DST respond to unforeseen external influences, such as international conflict, widespead drought? Are AI-DST-based decisions effective across multiple scales – benefitting individual farms and networked farms alike? Does the AI-DST promote use within farmer networks, community of practise and farmer cooperation? For example, AI-DSTs informed by shared data resources amongst farms within the same catchment. How do on-farm decisions influence the food supply chain – from farm to fork? Does the AI-DST provide this broader picture functionality? For example, does the AI-DST provide options, not only for farm productivity and farm emissions but also those concerned with externalities such as energy use, transportational costs, societal effects and more? What can be learnt and transferred from AI-DSTs and non-AI DSTs in other domains? Where appropriate, some of the above questions are illustrated using the unique and open datasets of four instrumented research farms at Rothamsted Research’s North Wyke Farm Platform (NWFP) in south west UK. The NWFP was established in 2010 to facilitate system-scale research (Takahashi et al. 2018), where to date over 400 in-situ sensors have been deployed and over 100 million measurements have been captured. Currently, the NWFP consists of two grassland (cattle and sheep), one arable and one indoor cattle farm.

Analysis of recent studies and publications. Recent research related to cloud systems of-fers computing paradigms for solving a large number of tasks of various types with minimal latency. The proposed paradigms can use resources, devices, nodes and clusters of the cloud center. But the main task of creating such paradigms is to solve the problem of offloading the cloud computing system using a combination of existing and new approaches. For example, when studying the cloud architecture, which is proposed in the form of a hierarchical boundary fog system [1], it is proposed to pay attention to the shift between the levels of the system according to the user's tolerance for latency. Many works of scientists have paid attention to the problem of using machine learning methods to solve the problems of offloading calculations and managing the mobility of the cloud system at different levels of its architecture. The current direction of cloud computing development is to solve the problem of load forecasting (proactive optimization) with the integration of intelligent agents into the system for monitoring, managing and adapting cloud service resources in real time, which interact with each other and central orchestrators for autonomous scaling and self-healing of the system [2].To study a more complex structure of the cloud computing system, the author's idea of the work [3] was used regarding the proposed strategy of dynamic joint unloading of cloud edge devices taking into account load balancing.Purpose of research. The purpose of the study is to consider a strategy for dynamic joint offloading of cloud devices, taking into account load balancing on edge servers and connection balancing on fog nodes for a multi-tier structure of a cloud computing system. Conducting experiments and analyzing the results. Presentation of the main research material. The author of the work [3] presented an ar-chitectural solution of a cloud system with the levels of end devices (DL), edge servers (EL) and cloud center (CL). In the proposed study, a fog level (FL) was added to approximate the model of the cloud computing system to the real one. A typical workflow of a cloud computing system is formed from several stages: users create tasks, end devices of the DL level distribute the received tasks to edge servers of the EL level through the main network under the control of the DL level offloading scheme, edge servers place the received tasks in a queue, on each edge server part of the tasks is processed locally, and the other part can be transferred to other edge servers for load balancing at the EL level; in parallel, tasks can be redirected due to the operation of the control scheme for selecting connection routers of the FL level; Part of the FL level tasks can also be redirected through the work of the cloud center for remote assistance under the control of the cloud offload scheme, after processing the tasks are delivered to the end devices to users from edge servers, fog nodes or from the cloud center.Conclusions. As a result of the study, a model of the evolution of the load of a multi-tier system was obtained to analyze the impact of the selected strategies for solving the problem of unloading the system levels. A model of a multi-tier system was proposed that combines the principles of edge, fog and cloud computing design, in which, when devices at the system levels are overloaded, part of the tasks can be partially redistributed.In the future, a hybrid system can be considered that includes a typical edge, fog and cloud computing system to study the issue of load optimization by transferring part of the tasks from edge servers and fog nodes directly to the cloud center or from edge servers, fog nodes, cloud center to the load control center.