Federated learning is a distributed deep learning method that trains models without sending local private training data to a server, achieving communication efficiency and security. However, current federated learning techniques have two key issues: 1) scalability limitations due to heavy traffic concentrated on the central server and 2) performance degradation caused by systems and data heterogeneity. To address these issues, we propose P2P-Fed, a decentralized federated learning approach with asynchronous aggregation to minimize delays and enhance concurrency for faster convergence. We adopt Chord, a popular distributed hash table protocol, to reduce the load on individual nodes in large-scale environments and ensure stable performance in high-churn networks. We designed and implemented this system to run in a real distributed environment. Experiments in various realistic scenarios demonstrated that P2P-Fed achieved up to a 6.9% performance improvement compared to the best-performing baseline algorithm, without incurring additional overhead.

Federated Learning (FL) enables the training of a single global model in a distributed manner while preserving the privacy of local data. In real-world FL environments, however, characterized by high heterogeneity in client data distributions (non-IID), the global model often suffers from performance degradation. To address this challenge, Personalized Federated Learning (PFL) has been proposed to provide personalized models either by clustering clients or by decoupling the model into shared and personalized components. Although partially effective, PFL using a client-server architecture often fails to capture client-specific characteristics and suffers from communication and computation bottlenecks. In this work, we present P3P-Fed, a personalized decentralized federated learning framework that performs clustering at the node level. By grouping nodes that have similar data distributions, each client achieves robust personalization even in highly heterogeneous non-IID environments where clustering at the global level is not feasible. To balance global model generalization and local personalization, P3P-Fed introduces a personalized head training strategy with partial aggregation and a differential learning rate. We evaluated P3P-Fed on a real-world testbed across varying degrees of data heterogeneity. The results show that P3P-Fed outperformed other baseline methods by as much as 8.64% and achieved the fastest convergence.

Non-volatile memory (NVM) provides bulk storage capacity, like NAND flash, while providing low latency, like DRAM, at the same time. NVM enables high-performance, reliable, and cost-effective high performance systems by providing low-latency data access and high capacity storage compared to traditional disk-based system. As NVM becomes a novel tier in the memory hierarchy, efficiently utilizing NVM I/O capability is important. In this work, we evaluate the I/O performance of NVM in three aspects: the performance change with a varying number of concurrent accesses, the performance dif-ference between remote and local accesses, and the performance change with various access granularity. We also compare the performance of NVM file systems that handle the different I/O characteristics of NVM. Specifically, Odinfs is the state-of-the-art NVM file system that solves the performance degradation of NVM with large number of threads and remote NUMA node accesses. We further optimize Odinfs by solving the I/O performance degradation with a small number of threads. We evaluate the optimized version of Odinfs and show that the throughput of Odinfs is increased by 30.91 % with four or fewer threads.

In HPC systems, the increasing need for a higher level of concurrency has led to packing more cores within a single chip. However, since multiple processes share memory space, the frequent access to resources in critical sections where only atomic operation has to be executed can result in poor performance. In this paper, we focus on reducing lock contention on the memory management system of an HPC manycore architecture. One of the critical sections causing severe lock contention in the I/O path is in the page management system, which uses multiple Least Recently Used (LRU) lists with a single lock instance. To solve this problem, we propose a Finer-LRU scheme, which optimizes the page reclamation process by splitting LRU lists into multiple sub-lists, each having its own lock instance. Our evaluation result shows that the Finer-LRU scheme can improve sequential write throughput by 57.03% and reduce latency by 98.94% compared to the baseline Linux kernel version 5.2.8 in the Intel Knights Landing (KNL) architecture.

Abstract Password-based authenticated key exchange (PAKE) allows two parties to compute a common secret key. PAKE offers the advantage of allowing two parties to pre-share only a password. However, when it is executed in a client–server environment, server corruption can expose the clients’ passwords. To be resilient against server compromises, verifier-based authenticated key exchange (VPAKE) is proposed, as an augmented version of PAKE. Thus far, there are two known major VPAKE constructions formally proven secure. However, both involve strong assumptions, such as random oracles. In this paper, we propose a simple and efficient VPAKE using tamper-proof hardware without random oracles to support resilient infrastructures. In particular, we transform Katz–Vaikuntanathan one-round PAKE into two-round VPAKE so as to instill resilience to server compromises. We provide a formal definition of VPAKE using tamper-proof hardware and security proof without random oracles. Finally, we provide a performance analysis and comparisons to previous VPAKE and PAKE protocols. Our transformation supports an efficient VPAKE protocol with six group element communications when the underlying Katz–Vaikuntanathan PAKE is instantiated by Cramer–Shoup ciphertext following the proposal by Benhamouda et al.

Burst buffers tolerate I/O spikes in High-Performance Computing environments by using a non-volatile flash technology. Burst buffers are commonly located between parallel file systems and compute nodes, handling bursty I/Os in the middle. In this architecture, burst buffers are shared resources. The performance of an SSD is significantly reduced when it is used excessively because of garbage collection, and we have observed that SSDs in a burst buffer become slow when many users simultaneously use the burst buffer. To mitigate the performance problem, we propose a new user-level I/O isolation framework in a High-Performance Computing environment using a multi-streamed SSD. The multi-streamed SSD allocates the same flash block for I/Os in the same stream. We assign a different stream to each user; thus, the user can use the stream exclusively. To evaluate the performance, we have used open-source supercomputing workloads and I/O traces from real workloads in the Cori supercomputer at the National Energy Research Scientific Computing Center. Via user-level I/O isolation, we have obtained up to a 125% performance improvement in terms of I/O throughput. In addition, our approach reduces the write amplification in the SSDs, leading to improved SSD endurance. This user-level I/O isolation framework could be applied to deployed burst buffers without having to make any user interface changes.

Solid State Drive (SSD) cost-per-bit continues to decrease. Consequently, system architects increasingly consider replacing Hard Disk Drives (HDDs) with SSDs to accelerate Hadoop MapReduce processing. When attempting this, system architects usually realize that SSD characteristics and today's Hadoop framework exhibit mismatches that impede indiscriminate SSD integration. Hence, cost-effective SSD utilization has proved challenging within many Hadoop environments. This paper compares SSD performance to HDD performance within a Hadoop MapReduce framework. It identifies extensible best practices that can exploit SSD benefits within Hadoop frameworks when combined with high network bandwidth and increased parallel storage access. Terasort benchmark results demonstrate that SSDs presently deliver significant cost-effectiveness when they store intermediate Hadoop data, leaving HDDs to store Hadoop Distributed File System (HDFS) source data.

Smart Grid applications are imposing challenging requirements of security and reliability on the N-way communication infrastructure being designed to support multiple grid applications. These challenges stem from the increasing incorporation of distributed renewable energy sources on to the grid, the rising deployment of electric vehicles, and active consumer participation into power grid operations, all of which communicate with the utility control center with varying degrees of priority and security. To address these challenging requirements, we propose SeDAX, a SEcure Data-centric Application eXtensible platform for Smart Grid applications. SeDAX implements scalable, resilient and secure data delivery and data sharing in a wide area network. The platform can scalably handle high volumes of data generated by both applications and sensors. The SeDAX architecture has as its basis a Delaunay Triangulation (DT) network. The properties of the DT graph are leveraged to scalably support secure data-centric (or information-centric) group communication. The primary goals of this platform are to support communication resilience and data availability. The key functional blocks of the SeDAX platform are: (1) a geographic hash forwarding algorithm that operates over the DT graph (DT-GHF), and (2) a DT-based data replication scheme. The forwarding and replication schemes are scalable and cost effective in terms of communication overhead and memory. We describe the design details of the SeDAX platform and present empirical results on the performance of SeDAX as compared with other geometric-based alternatives such as Geographic Hash Table (GHT) forwarding and Content Addressable Networking (CAN). The operation of SeDAX is illustrated in the context of implementing demand response, a known Smart Grid application.

We propose resource discovery and load balancing techniques to accommodate computing nodes with many types of computing elements, such as multi-core CPUs and GPUs, in a peer-to-peer desktop grid architecture. Heterogeneous nodes can have multiple types of computing elements, and the performance and characteristics of each computing element can be very different. Our scheme takes into account these diverse aspects of heterogeneous nodes to maximize overall system throughput. However, straightforward methods of handling diverse computing elements that differ on many axes can result in high overheads, both in local state and in communication volume. We describe approaches that minimize messaging costs without sacrificing the failure resilience provided by an underlying peer-to-peer overlay network. Simulation results show that our scheme's load balancing performance is comparable to that of a centralized approach, that communication costs are reduced significantly compared to the existing system, and that failure resilience is not compromised.

The majority of CPUs now sold contain multiple computing cores. However, current desktop grid computing systems either ignore the multiplicity of cores, or treat them as distinct, independent machines. The latter approach ignores the resource contention present between cores in a single CPU, while the former approach fails to take advantage of significant computing power. We propose a decentralized resource management framework for exploiting multi-core nodes in peer-to-peer grids. We present two new load-balancing schemes that explicitly account for the resource sharing and contention of multiple cores, and propose a simple simulation model that can represent a continuum of resource sharing among cores of a CPU. We use simulation to confirm that our two algorithms match jobs w ith computing nodes efficiently, and balance load during the lifetime of the computing jobs.

We present some recent and ongoing work in our decentralized desktop computing grid project. Specifically, we discuss matching jobs with compute nodes in a peer-to-peer grid of heterogeneous platforms, and the implementation of our algorithms in a concrete system.

In Korea, a large-scale IEEE 802.11b-based public wireless LAN (WLAN) service, called NESPOT, has been in operation by Korea Telecom (KT) across the country during the last four years. Along with a fast growth of the service, however, the service quality problems have been encountered. In order to manage and overcome such problems properly, we have developed a diagnostic tool, which is composed of a database for wireless connection-related dissatisfaction, a log collector and analyzer, and a mechanism to report breakdown of access points (APs). To our best knowledge, this is the first attempt to address an integrated management and develop a diagnosis architecture for large-scale commercial WLANs

Today's IEEE 802.11 Wireless LAN (WLAN) is an excellent solution for the broadband wireless networking. However, it lacks of the capability to support real-time services such as voice-over-IP (VoIP) properly. In this paper, we present a simple and viable approach to enhance the VoIP performance over the 802.11 WLAN by implementing two queues along with a strict priority queuing on top of the 802.11 medium access control (MAC) controller, e.g., in the device driver of the 802.11 cards. We find via extensive simulations that the proposed scheme is remarkably effective for the VoIP service in the infrastructure-based WLAN in the coexistence with the nonreal-time traffic thanks to the flow control mechanism of the TCP protocol, which is typically used for the nonreal-time traffic today. Due to its simplicity, the proposed scheme should be readily deployable in the existing WLANs via simple software upgrades for the enhanced VoIP services.