Source author record

Harel Levin

Harel Levin appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Distributed, Parallel, and Cluster Computing

Catalog footprint

What is connected

2works

1topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

Recovery of Distributed Iterative Solvers for Linear Systems Using Non-Volatile RAM

HPC systems are a critical resource for scientific research. The increased demand for computational power and memory ushers in the exascale era, in which supercomputers are designed to provide enormous computing power to meet these needs. These complex supercomputers consist of numerous compute nodes and are consequently expected to experience frequent faults and crashes. Mathematical solvers, in particular, iterative linear solvers are key building block in numerous large-scale scientific applications. Consequently, supporting the recovery of distributed solvers is necessary for scaling scientific applications to exascale platforms. Previous recovery methods for iterative solvers are based on Checkpoint-Restart (CR), which incurs high fault tolerance overhead, or intrinsic fault tolerance, which require extra computation time to converge after failures. Exact state reconstruction (ESR) was proposed as an alternative mechanism to alleviate the impact of frequent failures on long-term computations. ESR has been shown to provide exact reconstruction of the computation state while avoiding the need for costly checkpointing. However, ESR currently relies on volatile memory for fault tolerance, and must therefore maintain redundancies in the RAM of multiple nodes, incurring high memory and network overheads. Recent supercomputer designs feature emerging non-volatile RAM (NVRAM) technology. This paper investigates how NVRAM can be utilized to devise an enhanced ESR-based recovery mechanism that is more efficient and provides full resilience. Our mechanism, called in-NVRAM ESR, is based on a novel MPI One-Sided Communication (OSC) over RDMA implementation, and provides full resiliency while significantly reducing both the memory footprint and the time overhead in comparison with the original ESR design (in-RAM ESR).

preprint2021arXiv

Secured Distributed Algorithms Without Hardness Assumptions

We study algorithms in the distributed message-passing model that produce secured output, for an input graph $G$. Specifically, each vertex computes its part in the output, the entire output is correct, but each vertex cannot discover the output of other vertices, with a certain probability. This is motivated by high-performance processors that are embedded nowadays in a large variety of devices. In such situations, it no longer makes sense, and in many cases it is not feasible, to leave the whole processing task to a single computer or even a group of central computers. As the extensive research in the distributed algorithms field yielded efficient decentralized algorithms for many classic problems, the discussion about the security of distributed algorithms was somewhat neglected. Nevertheless, many protocols and algorithms were devised in the research area of secure multi-party computation problem (MPC or SMC). However, the notions and terminology of these protocols are quite different than in classic distributed algorithms. As a consequence, the focus in those protocols was to work for every function $f$ at the expense of increasing the round complexity, or the necessity of several computational assumptions. In this work, we present a novel approach, which rather than turning existing algorithms into secure ones, identifies and develops those algorithms that are inherently secure (which means they do not require any further constructions). This approach yields efficient secure algorithms for various locality problems, such as coloring, network decomposition, forest decomposition, and a variety of additional labeling problems. Remarkably, our approach does not require any hardness assumption, but only a private randomness generator in each vertex. This is in contrast to previously known techniques in this setting that are based on public-key encryption schemes.