DXT: Darshan eXtended Tracing
Explore the source record for details and available documents.
Engineering topics
Publications and source records attributed to Carns, Philip.
Explore the source record for details and available documents.
Contemporary high-performance computing (HPC) applications encompass a broad range of distinct I/O strategies and are often executed on a number of different compute platforms in their lifetime. These large-scale HPC platforms employ increasingly complex I/O subsystems to provide a suitable level of I/O performance to applications. Tuning I/O workloads for such a system is nontrivial, and the results generally are not portable to other HPC systems. I/O profiling tools can help to address this challenge, but most existing tools only instrument specific components within the I/O subsystem that provide a limited perspective on I/O performance. The increasing diversity of scientific applications and computing platforms calls for greater flexibililty and scope in I/O characterization.
Parallel discrete-event simulation (PDES) is an important tool in the codesign of extreme-scale systems because PDES provides a cost-effective way to evaluate designs of highperformance computing systems. Optimistic synchronization algorithms for PDES, such as Time Warp, allow events to be processed without global synchronization among the processing elements. A rollback mechanism is provided when events are processed out of timestamp order. Although optimistic synchronization protocols enable the scalability of large-scale PDES, the performance of the simulations must be tuned to reduce the number of rollbacks and provide an improved simulation runtime. To enable efficient large-scale optimistic simulations, one has to gain insight into the factors that affect the rollback behavior and simulation performance. We developed a tool for ROSS model developers that gives them detailed metrics on the performance of their large-scale optimistic simulations at varying levels of simulation granularity. Model developers can use this information for parameter tuning of optimistic simulations in order to achieve better runtime and fewer rollbacks. In this work, we instrument the ROSS optimistic PDES framework to gather detailed statistics about the simulation engine. We have also developed an interactive visualization interface that uses the data collected by the ROSS instrumentation to understand the underlying behavior of the simulation engine. The interface connects real time to virtual time in the simulation and provides the ability to view simulation data at different granularities. We demonstrate the usefulness of our framework by performing a visual analysis of the dragonfly network topology model provided by the CODES simulation framework built on top of ROSS. The instrumentation needs to minimize overhead in order to accurately collect data about the simulation performance. To ensure that the instrumentation does not introduce unnecessary overhead, we perform a scaling study that compares instrumented ROSS simulations with their noninstrumented counterparts in order to determine the amount of perturbation when running at different simulation scales.
Explore the source record for details and available documents.
Rich metadata in high-performance computing (HPC) systems contains extended information about users, jobs, data files, and their relationships. Property graphs are a promising data model to represent heterogeneous rich metadata flexibly. Specifically, a property graph can use vertices to represent different entities and edges to record the relationships between vertices with unique annotations. The high-volume HPC use case, with millions of entities and relationships, naturally requires an out-of-core distributed property graph database, which must support live updates (to ingest production information in real time), low-latency point queries (for frequent metadata operations such as permission checking), and large-scale traversals (for provenance data mining). Among these needs, large-scale property graph traversals are particularly challenging for distributed graph storage systems. Most existing graph systems implement a "level synchronous" breadth-first search algorithm that relies on global synchronization in each traversal step. This performs well in many problem domains; but a rich metadata management system is characterized by imbalanced graphs, long traversal lengths, and concurrent workloads, each of which has the potential to introduce or exacerbate stragglers (i.e., abnormally slow steps or servers in a graph traversal) that lead to low overall throughput for synchronous traversal algorithms. Previous research indicated that the straggler problem can be mitigated by using asynchronous traversal algorithms, and many graph-processing frameworks have successfully demonstrated this approach. Such systems require the graph to be loaded into a separate batch-processing framework instead of being iteratively accessed, however. In this work, we investigate a general asynchronous graph traversal engine that can operate atop a rich metadata graph in its native format. We outline a traversal-aware query language and key optimizations (traversal-affiliate caching and execution merging) necessary for efficient performance. We further explore the effect of different graph partitioning strategies on the traversal performance for both synchronous and asynchronous traversal engines. Our experiments show that the asynchronous graph traversal engine is more efficient than its synchronous counterpart in the case of HPC rich metadata processing, where more servers are involved and larger traversals are needed. Furthermore, the asynchronous traversal engine is more adaptive to different graph partitioning strategies.
Explore the source record for details and available documents.
As supercomputers close in on exascale performance, the increased number of processors and processing power translates to an increased demand on the underlying network interconnect. The Slim Fly network topology, a new lowdiameter and low-latency interconnection network, is gaining interest as one possible solution for next-generation supercomputing interconnect systems. In this paper, we present a high-fidelity Slim Fly it-level model leveraging the Rensselaer Optimistic Simulation System (ROSS) and Co-Design of Exascale Storage (CODES) frameworks. We validate our Slim Fly model with the Kathareios et al. Slim Fly model results provided at moderately sized network scales. We further scale the model size up to n unprecedented 1 million compute nodes; and through visualization of network simulation metrics such as link bandwidth, packet latency, and port occupancy, we get an insight into the network behavior at the million-node scale. We also show linear strong scaling of the Slim Fly model on an Intel cluster achieving a peak event rate of 36 million events per second using 128 MPI tasks to process 7 billion events. Detailed analysis of the underlying discrete-event simulation performance shows that a million-node Slim Fly model simulation can execute in 198 seconds on the Intel cluster.
Distributed object storage architectures have become the de facto standard for high-performance storage in big data, cloud, and HPC computing. Object storage deployments using commodity hardware to reduce costs often employ object replication as a method to achieve data resilience. Repairing object replicas after failure is a daunting task for systems with thousands of servers and billions of objects, however, and it is increasingly difficult to evaluate such scenarios at scale on realworld systems. Resilience and availability are both compromised if objects are not repaired in a timely manner. In this work we leverage a high-fidelity discrete-event simulation model to investigate replica reconstruction on large-scale object storage systems with thousands of servers, billions of objects, and petabytes of data. We evaluate the behavior of CRUSH, a well-known object placement algorithm, and identify configuration scenarios in which aggregate rebuild performance is constrained by object placement policies. After determining the root cause of this bottleneck, we then propose enhancements to CRUSH and the usage policies atop it to enable scalable replica reconstruction. We use these methods to demonstrate a simulated aggregate rebuild rate of 410 GiB/s (within 5% of projected ideal linear scaling) on a 1,024-node commodity storage system. We also uncover an unexpected phenomenon in rebuild performance based on the characteristics of the data stored on the system.
Here, with the increasing complexity of today’s high-performance computing (HPC) architectures, simulation has become an indispensable tool for exploring the design space of HPC systems—in particular, networks. In order to make effective design decisions, simulations of these systems must possess the following properties: (1) have high accuracy and fidelity, (2) produce results in a timely manner, and (3) be able to analyze a broad range of network workloads. Most state-of-the-art HPC network simulation frameworks, however, are constrained in one or more of these areas. In this work, we present a simulation framework for modeling two important classes of networks used in today’s IBM and Cray supercomputers: torus and dragonfly networks. We use the Co-Design of Multi-layer Exascale Storage Architecture (CODES) simulation framework to simulate these network topologies at a flit-level detail using the Rensselaer Optimistic Simulation System (ROSS) for parallel discrete-event simulation. Our simulation framework meets all the requirements of a practical network simulation and can assist network designers in design space exploration. First, it uses validated and detailed flit-level network models to provide an accurate and high-fidelity network simulation. Second, instead of relying on serial time-stepped or traditional conservative discrete-event simulations that limit simulation scalability and efficiency, we use the optimistic event-scheduling capability of ROSS to achieve efficient and scalable HPC network simulations on today’s high-performance cluster systems. Third, our models give network designers a choice in simulating a broad range of network workloads, including HPC application workloads using detailed network traces, an ability that is rarely offered in parallel with high-fidelity network simulations
Abstract—Darshan is a lightweight I/O characterization tool used to gather and summarize salient I/O workload statistics from HPC applications. Darshan was designed to minimize any possible perturbations of an application’s performance, leading it to be enabled by default on a number of production HPC systems. For each file accessed by a given application, Darshan records the count and types of I/O operations performed, histograms of access sizes, cumulative timers on the amount of time spent doing I/O, and other statistical data. This type of data has proved invaluable in understanding and improving the I/O performance of HPC applications. Darshan 3.0.0 is the new modularized version of the traditional Darshan library and file format, allowing users to easily add more in-depth I/O characterization data to Darshan logs. In this work we perform an empirical evaluation of Darshan 3.0.0 to ensure that it continues to meet performance expectations for broad deployment. In particular, we evaluate the imposed overhead on instrumented I/O operations, time taken to shut down and generate corresponding Darshan log files, and resultant log file sizes for different workloads. These performance results are compared to results of Darshan 2.3.0 on the Edison XC30 system at NERSC to determine whether the new version is lightweight enough to run full-time on production HPC systems. Our evaluation shows that Darshan has limited impact on application I/O performance and can fully generate a corresponding log file for most application workloads in under two seconds.
High-performance computing (HPC) systems face increasingly critical metadata management challenges, especially in the approaching exascale era. These challenges arise not only from exploding metadata volumes, but also from increasingly diverse metadata, which contains data provenance and arbitrary user-defined attributes in addition to traditional POSIX metadata. This ‘rich’ metadata is becoming critical to supporting advanced data management functionality such as data auditing and validation. In our prior work, we identified a graph-based model as a promising solution to uniformly manage HPC rich metadata due to its flexibility and generality. However, at the same time, graph-based HPC rich metadata anagement also introduces significant challenges to the underlying infrastructure. In this study, we first identify the challenges on the underlying infrastructure to support scalable, high-performance rich metadata management. Based on that, we introduce GraphMeta, a graphbased engine designed for this use case. It achieves performance scalability by introducing a new graph partitioning algorithm and a write-optimal storage engine. We evaluate GraphMeta under both synthetic and real HPC metadata workloads, compare it with other approaches, and demonstrate its advantages in terms of efficiency and usability for rich metadata management in HPC systems.
Specialized, transient data services are playing an increasingly prominent role in data-intensive scientific computing. These services offer flexible, on-demand pairing of applications with storage hardware using semantics that are optimized for the problem domain. Concurrent with this trend, upcoming scientific computing and big data systems will be deployed with emerging NVM technology to achieve the highest possible price/productivity ratio. Clearly, therefore, we must develop techniques to facilitate the confluence of specialized data services and NVM technology. In this work we explore how to enable the composition of NVM resources within transient distributed services while still retaining their essential performance characteristics. Our approach involves eschewing the conventional distributed file system model and instead projecting NVM devices as remote microservices that leverage user-level threads, RPC services, RMA-enabled network transports, and persistent memory libraries in order to maximize performance. We describe a prototype system that incorporates these concepts, evaluate its performance for key workloads on an exemplar system, and discuss how the system can be leveraged as a component of future data-intensive architectures.
Explore the source record for details and available documents.
Today’s massively parallel simulation codes can produce output ranging up to many terabytes of data. Utilizing this data to support scientific inquiry requires analysis and visualization, yet the sheer size of the data makes it cumbersome or impossible to read without computational resources similar to the original simulation. We identify two broad classes of problems for reading data and present effective solutions for both. The first class of data reads depends on user requirements and available resources. Tasks such as visualization and user-guided analysis may be accomplished using only a subset of variables with a restricted spatial extent at a reduced resolution. The other class of reads requires full resolution multivariate data to be loaded, for example to restart a simulation. We show that utilizing the hierarchical multiresolution IDX data format enables scalable and efficient serial and parallel read access on a variety of hardware from supercomputers down to portable devices. Here, we demonstrate interactive view-dependent visualization and analysis of massive scientific datasets using low-power commodity hardware, and we compare read performance with other parallel file formats for both full and partial resolution data.
Darshan is a production-quality I/O characteriza tion tool that captures and summarizes the I/O behavior of parallel applications. It records a variety of information with minimal overhead, including access patterns, number of files accessed, and the amount of time consumed by I/O routines. Darshan’s lightweight design makes it suitable for full-time deployment for workload characterization of large HPC systems.
High-performance computing (HPC) storage systems rely on access coordination to ensure that concurrent updates do not produce incoherent results. HPC storage systems typically employ pessimistic distributed locking to provide this functionality in cases where applications cannot perform their own coordination. This approach, however, introduces significant performance overhead and complicates fault handling. In this work we evaluate the viability of optimistic conditional storage operations as an alternative to distributed locking in HPC storage systems. We investigate design strategies and compare the two approaches in a prototype object storage system using a parallel read/modify/write benchmark. Our prototype illustrates that conditional operations can be easily integrated into distributed object storage systems and can outperform standard coordination primitives for simple update workloads. Our experiments show that conditional updates can achieve over two orders of magnitude higher performance than pessimistic locking for some parallel read/modify/write workloads.
Explore the source record for details and available documents.