A longstanding performance challenge in datacenter-based applications is how to efficiently handle incoming client requests that spawn many very short (μs scale) jobs that must be handled with high throughput and low tail latency. When no assumptions are made about the duration of individual jobs, or even about the distribution of their durations, this requires blind scheduling with frequent and efficient preemption, which is not scalably supported for μs-level tasks. We present Tiny Quanta (TQ), a system that enables efficient blind scheduling of μs-level workloads. TQ performs fine-grained preemptive scheduling and does so with high performance via a novel combination of two mechanisms: forced multitasking and two-level scheduling. Evaluations with a wide variety of μs-level workloads show that TQ achieves low tail latency while sustaining 1.2x to 6.8x the throughput of prior blind scheduling systems.

The layered Internet architecture, while far from perfect, has provided a global and neutral platform for the development of a wide range of applications. However, this core architecture has been increasingly augmented with additional in-network functionality that improves the performance, security, and privacy of these applications. These additional in-network functions, which are typically implemented at the network edge, are consistent with the layering of the Internet architecture but deviate from two of the core tenets of the Internet: interconnection and end-to-end simplicity. In this paper, we propose an architecture for these edge networking services called the InterEdge that applies these two Internet tenets in a manner appropriate to edge services while not requiring changes to the underlying Internet architecture or infrastructure.

We propose logical memory pools, a memory disaggregation architecture for the emerging Compute Express Link (CXL) technology in datacenters. The key idea is to create a memory pool by carving out parts of the local memory in each server, rather than using a physical memory pool that is separate from servers. Logical pools provide significant benefits over physical pools, namely, lower cost, support for near-memory computing without extra hardware, and flexibility on designating whether memory is part of the memory pool or not. We demonstrate that logical pools can execute workloads that are unfeasible in physical pools, and that its faster access leads to better performance. Realizing logical memory pools poses five major challenges, which we believe can be overcome. Given the benefits of logical pools, we believe the CXL community should refocus efforts on logical, rather than physical memory pools.

Events that take 10s to 100s of ns like cache misses increasingly cause CPU stalls. However, hiding the latency of these events is challenging: hardware mechanisms suffer from the lack of flexibility, whereas prior software mechanisms fall short due to large overhead and limited event visibility. In this paper, we argue that with a combination of two emerging techniques - light-weight coroutines and sample-based profiling, hiding these events in software is within reach.

As memory requirements grow, and advances in memory technology slow, the availability of sufficient main memory is increasingly the bottleneck in large compute clusters. One solution to this is memory disaggregation, where jobs can remotely access memory on other servers, or far memory. This paper first presents faster swapping mechanisms and a far memory-aware cluster scheduler that make it possible to support far memory at rack scale. Then, it examines the conditions under which this use of far memory can increase job throughput. We find that while far memory is not a panacea, for memory-intensive workloads it can provide performance improvements on the order of 10% or more even without changing the total amount of memory available.

In this paper we propose an extension to RDMA, called Remote Memory Calls (RMCs), that allows applications to install a customized set of 1-sided RDMA operations. We then explain how RMCs can be implemented on the forthcoming generation of SmartNICs and discuss the resulting tradeoffs between RMCs, 1-sided and 2-sided RDMA operations.

We present FireSim, an open-source simulation platform that enables cycle-exact microarchitectural simulation of large scale-out clusters by combining FPGA-accelerated simulation of silicon-proven RTL designs with a scalable, distributed network simulation. Unlike prior FPGA-accelerated simulation tools, FireSim runs on Amazon EC2 F1, a public cloud FPGA platform, which greatly improves usability, provides elasticity, and lowers the cost of large-scale FPGA-based experiments. We describe the design and implementation of FireSim and show how it can provide sufficient performance to run modern applications at scale, to enable true hardware-software co-design. As an example, we demonstrate automatically generating and deploying a target cluster of 1,024 3.2 GHz quad-core server nodes, each with 16 GB of DRAM, interconnected by a 200 Gbit/s network with 2 microsecond latency, which simulates at a 3.4 MHz processor clock rate (less than 1,000x slowdown over real-time). In aggregate, this FireSim instantiation simulates 4,096 cores and 16 TB of memory, runs ~14 billion instructions per second, and harnesses 12.8 million dollars worth of FPGAs—at a total cost of only ~$100 per simulation hour to the user. We present several examples to show how FireSim can be used to explore various research directions in warehouse-scale machine design, including modeling networks with high-bandwidth and low-latency, integrating arbitrary RTL designs for a variety of commodity and specialized datacenter nodes, and modeling a variety of datacenter organizations, as well as reusing the scale-out FireSim infrastructure to enable fast, massively parallel cycle-exact single-node microarchitectural experimentation.

Approximate computing trades quality of application output for higher efficiency and performance. Approximation is useful only if its impact on application output quality is acceptable to the users. However, there is a lack of systematic solutions and studies that explore users' perspective on the effects of approximation. In this paper, we seek to provide one such solution for the developers to probe and discover the boundary of quality loss that most users will deem acceptable. We propose AxGames, a crowdsourced solution that enables developers to readily infer a statistical common ground from the general public through three entertaining games. The users engage in these games by betting on their opinion about the quality loss of the final output while the AxGames framework collects statistics about their perceptions. The framework then statistically analyzes the results to determine the acceptable levels of quality for a pair of (application, approximation technique). The three games are designed such that they effectively capture quality requirements with various tradeoffs and contexts. To evaluate AxGames, we examine seven diverse applications that produce user perceptible outputs and cover a wide range of domains, including image processing, optical character recognition, speech to text conversion, and audio processing. We recruit 700 participants/users through Amazon's Mechanical Turk to play the games that collect statistics about their perception on different levels of quality. Subsequently, the AxGames framework uses the Clopper-Pearson exact method, which computes a binomial proportion confidence interval, to analyze the collected statistics for each level of quality. Using this analysis, AxGames can statistically project the quality level that satisfies a given percentage of users. The developers can use these statistical projections to tune the level of approximation based on the user experience. We find that the level of acceptable quality loss significantly varies across applications. For instance, to satisfy 90% of users, the level of acceptable quality loss is 2% for one application (image processing) and 26% for another (audio processing). Moreover, the pattern with which the crowd responds to approximation takes significantly different shape and form depending on the class of applications. These results confirm the necessity of solutions that systematically explore the effect of approximation on the end user experience.

A growing number of commercial and enterprise systems increasingly rely on compute-intensive Machine Learning (ML) algorithms. While the demand for these compute-intensive applications is growing, the performance benefits from general-purpose platforms are diminishing. Field Programmable Gate Arrays (FPGAs) provide a promising path forward to accommodate the needs of machine learning algorithms and represent an intermediate point between the efficiency of ASICs and the programmability of general-purpose processors. However, acceleration with FPGAs still requires long development cycles and extensive expertise in hardware design. To tackle this challenge, instead of designing an accelerator for a machine learning algorithm, we present TABLA, a framework that generates accelerators for a class of machine learning algorithms. The key is to identify the commonalities across a wide range of machine learning algorithms and utilize this commonality to provide a high-level abstraction for programmers. TABLA leverages the insight that many learning algorithms can be expressed as a stochastic optimization problem. Therefore, learning becomes solving an optimization problem using stochastic gradient descent that minimizes an objective function over the training data. The gradient descent solver is fixed while the objective function changes for different learning algorithms. TABLA provides a template-based framework to accelerate this class of learning algorithms. Therefore, a developer can specify the learning task by only expressing the gradient of the objective function using our high-level language. Tabla then automatically generates the synthesizable implementation of the accelerator for FPGA realization using a set of hand-optimized templates. We use Tabla to generate accelerators for ten different learning tasks targeted at a Xilinx Zynq FPGA platform. We rigorously compare the benefits of FPGA acceleration to multi-core CPUs (ARM Cortex A15 and Xeon E3) and many-core GPUs (Tegra K1, GTX 650 Ti, and Tesla K40) using real hardware measurements. TABLA-generated accelerators provide 19.4x and 2.9x average speedup over the ARM and Xeon processors, respectively. These accelerators provide 17.57x, 20.2x, and 33.4x higher Performance-per-Watt in comparison to Tegra, GTX 650 Ti and Tesla, respectively. These benefits are achieved while the programmers write less than 50 lines of code.

Deep Neural Networks (DNNs) are compute-intensive learning models with growing applicability in a wide range of domains. FPGAs are an attractive choice for DNNs since they offer a programmable substrate for acceleration and are becoming available across different market segments. However, obtaining both performance and energy efficiency with FPGAs is a laborious task even for expert hardware designers. Furthermore, the large memory footprint of DNNs, coupled with the FPGAs' limited on-chip storage makes DNN acceleration using FPGAs more challenging. This work tackles these challenges by devising DnnWeaver, a framework that automatically generates a synthesizable accelerator for a given (DNN, FPGA) pair from a high-level specification in Caffe [1]. To achieve large benefits while preserving automation, DNNWEAVER generates accelerators using hand-optimized design templates. First, DnnWeaver translates a given high-level DNN specification to its novel ISA that represents a macro dataflow graph of the DNN. The DnnWeaver compiler is equipped with our optimization algorithm that tiles, schedules, and batches DNN operations to maximize data reuse and best utilize target FPGA's memory and other resources. The final result is a custom synthesizable accelerator that best matches the needs of the DNN while providing high performance and efficiency gains for the target FPGA. We use DnnWeaver to generate accelerators for a set of eight different DNN models and three different FPGAs, Xilinx Zynq, Altera Stratix V, and Altera Arria 10. We use hardware measurements to compare the generated accelerators to both multicore CPUs (ARM Cortex A15 and Xeon E3) and many-core GPUs (Tegra K1, GTX 650Ti, and Tesla K40). In comparison, the generated accelerators deliver superior performance and efficiency without requiring the programmers to participate in the arduous task of hardware design.

Cyclic instabilities can impact the performance of a multi agent system, especially in terms of the user's point of view. Different strategies can be used in order to prevent this problem. In this paper we present two strategies, ONL1 and ONL2 which aim at minimizing the collateral consequences of locking (shutting down the agent). These two strategies focus on minimizing the total number of locked nodes and the total sum of locked nodes' weights (importance). Both these strategies performed better than a former developed methodology - INPRES-, especially in very dense and coupled systems.