Toward Scalable and Asynchronous Object-Centric Data Management for HPC

Abstract

Emerging high performance computing (HPC) systems are expected to be deployed with an unprecedented level of complexity due to a deep system memory and storage hierarchy. Efficient and scalable methods of data management and movement through this hierarchy is critical for scientific applications using exascale systems. Moving toward new paradigms for scalable I/O in the extreme-scale era, we introduce novel object-centric data abstractions and storage mechanisms that take advantage of the deep storage hierarchy, named Proactive Data Containers (PDC). In this paper, we formulate object-centric PDCs and their mappings in different levels of the storage hierarchy. PDC adopts a client-server architecture with a set of servers managing data movement across storage layers. To demonstrate the effectiveness of the proposed PDC system, we have measured performance of benchmarks and I/O kernels from scientific simulation and analysis applications using PDC programming interface, and compared the results with existing highly tuned I/O libraries. Using asynchronous I/O along with data and metadata optimizations, PDC demonstrates up to 23× speedup over HDF5 and PLFS in writing and reading data from a plasma physics simulation. PDC achieves comparable performance with HDF5 and PLFS in reading and writing data of a single timestep at small scale, and outperforms them at a scale of larger than ten thousand cores. In contrast to existing storage systems, PDC offers user-space data management with the flexibility to allocate the number of PDC servers depending on the workload.

Interfacing HDF5 with a scalable object‐centric storage system on hierarchical storage

“…In this section, we provide an overview of the proactive data containers system and focus on the capabilities and semantics that it provides. For full details of the PDC system implementation, please refer to our previous publications …”

“…Object‐based storage systems provide semantics that have the potential to reduce the complexity of storage systems, as well as to improve performance. We have developed a user‐space object‐centric storage and data management system, called proactive data containers (PDC) . PDC provides scalable distributed metadata management and the capabilities of transparent, asynchronous, and autonomous data movement to multiple storage tiers .…”

“…We have developed a user‐space object‐centric storage and data management system, called proactive data containers (PDC) . PDC provides scalable distributed metadata management and the capabilities of transparent, asynchronous, and autonomous data movement to multiple storage tiers . PDC offers a programming interface that applications can use in order to take advantage of the data and metadata management services.…”

Dart

“…We implemented DART on the same software platform that has been used in SoMeta [43] and PDC [44], which is built upon MPI and Mercury [41] RPC framework. The software platform already contains DHT implementation which already facilitated the experiments in both aforementioned research works.…”

Demystifying asynchronous I/O Interference in HPC applications

“…Using this approach, applications can resume immediately after interacting with the local storage, which reduces the I/O overhead, thus improving the overall runtime performance and scalability. Such an approach is used in a wide range of data management services (Barisits et al, 2019; Tang et al, 2018, 2019), in-situ analytics (Pebay et al, 2016), checkpointing runtimes (Ferreira et al, 2014; Nicolae et al, 2019), and task-based runtimes (Augonnet et al, 2011; Duran et al, 2011; Kale and Krishnan, 1993).…”