Search

Single-Program-Multiple-Data (SPMD) parallelism has recently been adopted to train large deep neural networks (DNNs). Few studies have explored its applicability on heterogeneous clusters, to fully exploit available resources for large model learning. This paper presents \OurSystem, an automated system designed to expedite SPMD DNN training on heterogeneous clusters. \OurSystem jointly optimizes the tensor sharding strategy, sharding ratios across heterogeneous devices and the communication methods for tensor exchanges for optimized distributed training with SPMD parallelism. We novelly formulate model partitioning as a program synthesis problem, in which we generate a distributed program from scratch on a distributed instruction set that semantically resembles the program designed for a single device, and systematically explore the solution space with an A*-based search algorithm. We derive the optimal tensor sharding ratios by formulating it as a linear programming problem. Additionally, \OurSystem explores tensor communication optimization in a heterogeneous cluster and integrates it as part of the program synthesis process, for automatically choosing optimal collective communication primitives and applying sufficient factor broadcasting technique. Extensive experiments on representative workloads demonstrate that \OurSystem achieves up to 2.41x speed-up on heterogeneous clusters., Comment: EuroSys '24

Pipeline parallelism has been demonstrated to be a remarkable approach to improve throughput for training deep neural networks with billions of parameters over heterogeneous clusters. The 1F1B scheduling plan is a widely adopted strategy for memory and performance optimization, which interchanges the forward and backward stage computations of different micro-batches. On the other hand, a common issue in using the 1F1B scheduling is that stage computation is delayed due to the data transfer when network resources are preempted by other tasks, even with the minimum communication between stages. The exclusive access of these network resources cannot be guaranteed in cloud offerings. We present a general scheduling technique to accommodate pipeline parallelism to preempted network environments at the expense of a certain amount of memory pressure. The core concept is to extend 1F1B schedule scheme to kFkB, which groups k micro-batches, and alternately executes k forward and backward computations. We propose Ada-Grouper, an adaptive kFkB scheduler which regularly adjusts the number of group members k to maintain an optimal balance between communication and computation efficiency correspond to changes in a changing network environment under the memory limit. Experimental results demonstrate that our design maintain stable performance for pipeline parallelism, yielding a performance increase of up from 4% to 30%, compared with 1F1B in preempted network scenarios.

We present Rhino, a system for accelerating tensor programs with automatic parallelization on AI platform for real production environment. It transforms a tensor program written for a single device into an equivalent distributed program that is capable of scaling up to thousands of devices with no user configuration. Rhino firstly works on a semantically independent intermediate representation of tensor programs, which facilitates its generalization to unprecedented applications. Additionally, it implements a task-oriented controller and a distributed runtime for optimal performance. Rhino explores on a complete and systematic parallelization strategy space that comprises all the paradigms commonly employed in deep learning (DL), in addition to strided partitioning and pipeline parallelism on non-linear models. Aiming to efficiently search for a near-optimal parallel execution plan, our analysis of production clusters reveals general heuristics to speed up the strategy search. On top of it, two optimization levels are designed to offer users flexible trade-offs between the search time and strategy quality. Our experiments demonstrate that Rhino can not only re-discover the expert-crafted strategies of classic, research and production DL models, but also identify novel parallelization strategies which surpass existing systems for novel models.

It is a challenging task to train large DNN models on sophisticated GPU platforms with diversified interconnect capabilities. Recently, pipelined training has been proposed as an effective approach for improving device utilization. However, there are still several tricky issues to address: improving computing efficiency while ensuring convergence, and reducing memory usage without incurring additional computing costs. We propose DAPPLE, a synchronous training framework which combines data parallelism and pipeline parallelism for large DNN models. It features a novel parallelization strategy planner to solve the partition and placement problems, and explores the optimal hybrid strategy of data and pipeline parallelism. We also propose a new runtime scheduling algorithm to reduce device memory usage, which is orthogonal to re-computation approach and does not come at the expense of training throughput. Experiments show that DAPPLE planner consistently outperforms strategies generated by PipeDream's planner by up to 3.23x under synchronous training scenarios, and DAPPLE runtime outperforms GPipe by 1.6x speedup of training throughput and reduces the memory consumption of 12% at the same time.

We introduce a new code for cosmological simulations, PHoToNs, which has features on performing massive cosmological simulations on heterogeneous high performance Computer (HPC) and threads oriented programming. PHoToNs adopts a hybrid scheme to compute gravity force, with the conventional PM to compute the long-range force, the Tree algorithm to compute the short range force, and the direct summation PP to compute the gravity from very close particles. A self-similar space filling Peano-Hilbert curve is used to decompose computing domain. Threads programming is highly used to more flexibly manage the domain communication, PM calculation and synchronization, as well as Dual Tree Traversal on the CPU+MIC platform. The scalability of the PHoToNs performs well and the efficiency of PP kernel achieves 68.6% of peak performance on MIC and 74.4% on CPU platforms. We also test the accuracy of the code against the much used Gadget-2 in the community and found excellent agreement., Comment: Accepted for publication in RAA, 14 pages, 7 figures