Recommendation systems have been widely embedded into many Internet services. For example, Meta’s deep learning recommendation model (DLRM) shows high predictive accuracy of click-through rate in processing large-scale embedding tables. The SparseLengthSum (SLS) kernel of the DLRM dominates the inference time of the DLRM due to intensive irregular memory accesses to the embedding vectors. Some prior works directly adopt near-data processing (NDP) solutions to obtain higher memory bandwidth to accelerate SLS. However, their inferior memory hierarchy induces a low performance-cost ratio and fails to fully exploit the data locality. Although some software-managed cache policies were proposed to improve the cache hit rate, the incurred cache miss penalty is unacceptable considering the high overheads of executing the corresponding programs and the communication between the host and the accelerator. To address the issues aforementioned, we proposeEMS-i , an efficient memory system design that integrates Solid State Drive (SSD) into the memory hierarchy using Compute Express Link (CXL) for recommendation system inference. We specialize the caching mechanism according to the characteristics of various DLRM workloads and propose a novel prefetching mechanism to further improve the performance. In addition, we delicately design the inference kernel and develop a customized mapping scheme for SLS operation, considering the multi-level parallelism in SLS and the data locality within a batch of queries. Compared to the state-of-the-art NDP solutions,EMS-i achieves up to 10.9× speedup over RecSSD and the performance comparable to RecNMP with 72% energy savings.EMS-i also saves up to 8.7× and 6.6 × memory cost w.r.t. RecSSD and RecNMP, respectively.
more »
« less
This content will become publicly available on July 10, 2025
OPER: Optimality-Guided Embedding Table Parallelization for Large-scale Recommendation Model
The deployment of Deep Learning Recommendation Models (DLRMs) involves the parallelization of extra-large embedding tables (EMTs) on multiple GPUs. Existing works overlook the input-dependent behavior of EMTs and parallelize them in a coarse-grained manner, resulting in unbalanced workload distribution and inter-GPU communication.
To this end, we propose OPER, an algorithm-system co-design with OPtimality-guided Embedding table parallelization for large-scale Recommendation model training and inference. The core idea of OPER is to explore the connection between DLRM inputs and the efficiency of distributed EMTs, aiming to provide a near-optimal parallelization strategy for EMTs. Specifically, we conduct an in-depth analysis of various types of EMTs parallelism and propose a heuristic search algorithm to efficiently approximate an empirically near-optimal EMT parallelization. Furthermore, we implement a distributed shared memory-based system, which supports the lightweight but complex computation and communication pattern of fine-grained EMT parallelization, effectively converting theoretical improvements into real speedups. Extensive evaluation shows that OPER achieves 2.3× and 4.0× speedup on average in training and inference, respectively, over state-of-the-art DLRM frameworks.
more »
« less
- Award ID(s):
- 2124039
- PAR ID:
- 10538948
- Publisher / Repository:
- USENIX Association
- Date Published:
- ISBN:
- 978-1-939133-41-0
- Format(s):
- Medium: X
- Location:
- Santa Clara, CA, USA
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
We propose a new data structure called CachedEmbeddings for training large scale deep learning recommendation models (DLRM) efficiently on heterogeneous (DRAM + non-volatile) memory platforms. CachedEmbeddings implements an implicit software-managed cache and data movement optimization that is integrated with the Julia programming framework to optimize the implementation of large scale DLRM implementations with multiple sparse embedded tables operations. In particular we show an implementation that is 1.4X to 2X better than the best known Intel CPU based implementations on state-of-the-art DLRM benchmarks on a real heterogeneous memory platform from Intel, and 1.32X to 1.45X improvement over Intel’s 2LM implementation that treats the DRAM as a hardware managed cache.more » « less
-
Deep Learning Recommendation Models (DLRMs) are very popular in personalized recommendation systems and are a major contributor to the data-center AI cycles. Due to the high computational and memory bandwidth needs of DLRMs, specifically the embedding stage in DLRM inferences, both CPUs and GPUs are used for hosting such workloads. This is primarily because of the heavy irregular memory accesses in the embedding stage of computation that leads to significant stalls in the CPU pipeline. As the model and parameter sizes keep increasing with newer recommendation models, the computational dominance of the embedding stage also grows, thereby, bringing into question the suitability of CPUs for inference. In this paper, we first quantify the cause of irregular accesses and their impact on caches and observe that off-chip memory access is the main contributor to high latency. Therefore, we exploit two well-known techniques: (1) Software prefetching, to hide the memory access latency suffered by the demand loads and (2) Overlapping computation and memory accesses, to reduce CPU stalls via hyperthreading to minimize the overall execution time. We evaluate our work on a single-core and 24-core configuration with the latest recommendation models and recently released production traces. Our integrated techniques speed up the inference by up to 1.59x, and on average by 1.4x.more » « less
-
In this paper, we consider hybrid parallelism—a paradigm that em- ploys both Data Parallelism (DP) and Model Parallelism (MP)—to scale distributed training of large recommendation models. We propose a compression framework called Dynamic Communication Thresholding (DCT) for communication-efficient hybrid training. DCT filters the entities to be communicated across the network through a simple hard-thresholding function, allowing only the most relevant information to pass through. For communication efficient DP, DCT compresses the parameter gradients sent to the parameter server during model synchronization. The threshold is updated only once every few thousand iterations to reduce the computational overhead of compression. For communication efficient MP, DCT incorporates a novel technique to compress the activations and gradients sent across the network during the forward and backward propagation, respectively. This is done by identifying and updating only the most relevant neurons of the neural network for each training sample in the data. We evaluate DCT on publicly available natural language processing and recommender models and datasets, as well as recommendation systems used in production at Facebook. DCT reduces communication by at least 100× and 20× during DP and MP, respectively. The algorithm has been deployed in production, and it improves end-to-end training time for a state-of-the-art industrial recommender model by 37%, without any loss in performance.more » « less
-
We consider a hierarchical inference system with multiple clients connected to a server via a shared communication resource. When necessary, clients with low-accuracy machine learning models can offload classification tasks to a server for processing on a high-accuracy model. We propose a distributed online offloading algorithm which maximizes the accuracy subject to a shared resource utilization constraint thus indirectly realizing accuracy-delay tradeoffs possible given an underlying network scheduler. The proposed algorithm, named Lyapunov-EXP4, introduces a loss structure based on Lyapunov-drift minimization techniques to the bandits with expert advice framework. We prove that the algorithm converges to a near-optimal threshold policy on the confidence of the clients’ local inference without prior knowledge of the system’s statistics and efficiently solves a constrained bandit problem with sublinear regret. We further consider settings where clients may employ multiple thresholds, allowing more aggressive optimization of overall accuracy at a possible loss in fairness. Extensive simulation results on real and synthetic data demonstrate convergence of Lyapunov-EXP4, and show themore » « less
