An official website of the United States government
Transparently checkpointing MPI for fault tolerance and load balancing is a long-standing problem in HPC. The problem has been complicated by the need to provide checkpoint-restart services for all combinations of an MPI implementation over all network interconnects. This work presents MANA (MPI-Agnostic Network-Agnostic transparent checkpointing), a single code base which supports all MPI implementation and interconnect combinations. The agnostic properties imply that one can checkpoint an MPI application under one MPI implementation and perhaps over TCP, and then restart under a second MPI implementation over InfiniBand on a cluster with a different number of CPU cores per node. This technique is based on a novel "split-process" approach, which enables two separate programs to co-exist within a single process with a single address space. This work overcomes the limitations of the two most widely adopted transparent checkpointing solutions, BLCR and DMTCP/InfiniBand, which require separate modifications to each MPI implementation and/or underlying network API. The runtime overhead is found to be insignificant both for checkpoint-restart within a single host, and when comparing a local MPI computation that was migrated to a remote cluster against an ordinary MPI computation running natively on that same remote cluster.
more »
« less
Optimizing Asynchronous Multi-Level Checkpoint/Restart Configurations with Machine Learning
With the emergence of versatile storage systems, multi-level checkpointing (MLC) has become a common approach to gain efficiency. However, multi-level checkpoint/restart can cause enormous I/O traffic on HPC systems. To use multilevel checkpointing efficiently, it is important to optimize checkpoint/restart configurations. Current approaches, namely modeling and simulation, are either inaccurate or slow in determining the optimal configuration for a large scale system. In this paper, we show that machine learning models can be used in combination with accurate simulation to determine the optimal checkpoint configurations. We also demonstrate that more advanced techniques such as neural networks can further improve the performance in optimizing checkpoint configurations.
more »
« less
- PAR ID:
- 10156303
- Date Published:
- Journal Name:
- The IEEE International Workshop on High-Performance Storage
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
MANA-2.0 is a scalable, future-proof design for transparent checkpointing of MPI-based computations. Its network transparency (“network-agnostic”) feature ensures that MANA-2.0 will provide a viable, efficient mechanism for trans-parently checkpointing MPI applications on current and future supercomputers. MANA-2.0 is an enhancement of previous work, the original MANA, which interposes MPI calls, and is a work in progress intended for production deployment. MANA-2.0 implements a series of new algorithms and features that improve MANA's scalability and reliability, enabling transparent checkpoint-restart over thousands of MPI processes. MANA-2.0 is being tested on today's Cori supercomputer at NERSC using Cray MPICH library over the Cray GNI network, but it is designed to work over any standard MPI running over an arbitrary network. Two widely-used HPC applications were selected to demonstrate the enhanced features of MANA-2.0: GROMACS, a molecular dynamics simulation code with frequent point-to-point communication, and VASP, a materials science code with frequent MPI collective communication. Perhaps the most important lesson to be learned from MANA-2.0 is a series of algorithms and data structures for library-based transformations that enable MPI-based computations over MANA-2.0 to reliably survive the checkpoint-restart transition.more » « less
-
Unified Virtual Memory (UVM) was recently introduced with CUDA version 8 and the Pascal GPU. The older CUDA programming style is akin to older large-memory UNIX applications which used to directly load and unload memory segments. Newer CUDA programs have started taking advantage of UVM for the same reasons of superior programmability that UNIX applications long ago switched to assuming the presence of virtual memory. Therefore, checkpointing of UVM has become increasing important, especially as NVIDIA CUDA continues to gain wider popularity: 87 of the top 500 supercomputers in the latest listings use NVIDIA GPUs, with a current trend of ten additional NVIDIA-based supercomputers each year. A new scalable checkpointing mechanism, CRUM (Checkpoint-Restart for Unified Memory), is demonstrated for hybrid CUDA/MPI computations across multiple computer nodes. The support for UVM is particularly attractive for programs requiring more memory than resides on the GPU, since the alternative to UVM is for the application to directly copy memory between device and host. Furthermore, CRUM supports a fast, forked checkpointing, which mostly overlaps the CUDA computation with storage of the checkpoint image in stable storage. The runtime overhead of using CRUM is 6% on average, and the time for forked checkpointing is seen to be a factor of up to 40 times less than traditional, synchronous checkpointing.more » « less
-
Accepted and published in the Proceedings of the 2025 USENIX Annual Technical Conference (USENIX ATC ’25). Deep neural network (DNN) training continues to scale rapidly in terms of model size, data volume, and sequence length, to the point where multiple machines are required to fit large models for training. Different distributed and parallel training strategies have been developed to support large-scale DNN training by partitioning the training state across GPUs. However, existing DNN training systems provide very limited support for reconfiguring parallelism strategies in the middle of the training via checkpointing. This limitation arises because distributed checkpoints are tightly coupled to specific model parallelism and hardware configurations, preventing large-scale training jobs from efficiently adapting to hardware failures or resource elasticity. This paper presents Universal Checkpointing (UCP), a novel checkpointing system that enables flexible and efficient DNN training with reconfigurable parallelism. UCP overcomes challenges in existing systems by decoupling checkpoint structure from parallel training strategies and hardware configurations. In addition, we present a pattern-based reconfiguration pipeline that enables automatic, flexible, and efficient mapping of checkpoint state to various parallelism strategies. Evaluation on a range of DNN models, including state-of-the-art dense and sparse LLMs, shows that UCP enables reconfiguration for a broader set of widely used parallelism strategies than existing solutions while adding negligible reconfiguration cost. UCP has been successfully employed in real LLM training workloads, greatly enhancing their flexibility and resilience to dynamic hardware environments.more » « less
-
null (Ed.)The importance of fault tolerance continues to increase for HPC applications. The continued growth in size and complexity of HPC systems, and of the applications them- selves, is leading to an increased likelihood of failures during execution. However, most HPC programming models do not have a built-in fault tolerance mechanism. Instead, application developers usually rely on external support such as application- level checkpoint-restart (C/R) libraries to make their codes fault tolerant. However, this increases the burden on the application developer, who must use the libraries carefully to ensure correct behavior and to minimize the overheads. The C/R routines will be employed to save the values of all needed program variables at the places in the code where they are invoked. It is important for correctness that the program data is in a consistent state at these places. It is non-trivial to determine such points in OpenSHMEM, which relies upon single-sided communications to provide high performance. The amount of data to be collected, and the frequency with which this is performed, must also be carefully tuned, as the overheads introduced by C/R calls can be extremely high. There is very little prior work on checkpoint-restart support in the context of the OpenSHMEM programming interface. In this paper, we introduce OpenSHMEM and describe the challenges it poses for checkpointing. We identify the safest places for inserting C/R calls in an OpenSHMEM program and describe a straightforward approach for identifying the data that needs to be checkpointed at these positions in the code. We provide these two functionalities in a tool that exploits compiler analyses to propose checkpoints and the sets of data for saving at them, to the application developer.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
