arXiv subjects

Vijay Rao

Publications and source records attributed to Vijay Rao.

6 recordsLinked to original sources

arXiv2023

Software-Hardware Co-design for Fast and Scalable Training of Deep Learning Recommendation Models

Deep learning recommendation models (DLRMs) are used across many business-critical services at Facebook and are the single largest AI application in terms of infrastructure demand in its data-centers. In this paper we discuss the SW/HW co-designed solution for high-performance distributed training of large-scale DLRMs. We introduce a high-performance scalable software stack based on PyTorch and pair it with the new evolution of Zion platform, namely ZionEX. We demonstrate the capability to train very large DLRMs with up to 12 Trillion parameters and show that we can attain 40X speedup in terms of time to solution over previous systems. We achieve this by (i) designing the ZionEX platform with dedicated scale-out network, provisioned with high bandwidth, optimal topology and efficient transport (ii) implementing an optimized PyTorch-based training stack supporting both model and data parallelism (iii) developing sharding algorithms capable of hierarchical partitioning of the embedding tables along row, column dimensions and load balancing them across multiple workers; (iv) adding high-performance core operators while retaining flexibility to support optimizers with fully deterministic updates (v) leveraging reduced precision communications, multi-level memory hierarchy (HBM+DDR+SSD) and pipelining. Furthermore, we develop and briefly comment on distributed data ingestion and other supporting services that are required for the robust and efficient end-to-end training in production environments.

cs.DC↗

arXiv2021

Supporting Massive DLRM Inference Through Software Defined Memory

Deep Learning Recommendation Models (DLRM) are widespread, account for a considerable data center footprint, and grow by more than 1.5x per year. With model size soon to be in terabytes range, leveraging Storage ClassMemory (SCM) for inference enables lower power consumption and cost. This paper evaluates the major challenges in extending the memory hierarchy to SCM for DLRM, and presents different techniques to improve performance through a Software Defined Memory. We show how underlying technologies such as Nand Flash and 3DXP differentiate, and relate to real world scenarios, enabling from 5% to 29% power savings.

cs.AR↗

arXiv2021

Towards Enabling High-Five Over WiFi

The next frontier for immersive applications is enabling sentience over the Internet. Tactile Internet (TI) envisages transporting skills by providing Ultra-Low Latency (ULL) communications for transporting touch senses. In this work, we focus our study on the first/last mile communication, where the future generation WiFi-7 is pitched as the front-runner for ULL applications. We discuss a few candidate features of WiFi-7 and highlight its major pitfalls with respect to ULL communication. Further, through a specific implementation of WiFi-7 (vanilla WiFi-7) in our custom simulator, we demonstrate the impact of one of the pitfalls - standard practice of using jitter buffer in conjunction with frame aggregation - on TI communication. To circumvent this, we propose Non-Buffered Scheme (NoBuS) - a simple MAC layer enhancement for enabling TI applications on WiFi-7. NoBuS trades off packet loss for latency enabling swift synchronization between the master and controlled domains. Our findings reveal that employing NoBuS yields a significant improvement in RMSE of TI signals. Further, we show that the worst-case WiFi latency with NoBuS is 3.72 ms - an order of magnitude lower than vanilla WiFi-7 even under highly congested network conditions.

cs.NI↗

arXiv2019

Deep Learning Recommendation Model for Personalization and Recommendation Systems

With the advent of deep learning, neural network-based recommendation models have emerged as an important tool for tackling personalization and recommendation tasks. These networks differ significantly from other deep learning networks due to their need to handle categorical features and are not well studied or understood. In this paper, we develop a state-of-the-art deep learning recommendation model (DLRM) and provide its implementation in both PyTorch and Caffe2 frameworks. In addition, we design a specialized parallelization scheme utilizing model parallelism on the embedding tables to mitigate memory constraints while exploiting data parallelism to scale-out compute from the fully-connected layers. We compare DLRM against existing recommendation models and characterize its performance on the Big Basin AI platform, demonstrating its usefulness as a benchmark for future algorithmic experimentation and system co-design.

cs.IR↗

arXiv2018

Deep Learning Inference in Facebook Data Centers: Characterization, Performance Optimizations and Hardware Implications

The application of deep learning techniques resulted in remarkable improvement of machine learning models. In this paper provides detailed characterizations of deep learning models used in many Facebook social network services. We present computational characteristics of our models, describe high performance optimizations targeting existing systems, point out their limitations and make suggestions for the future general-purpose/accelerated inference hardware. Also, we highlight the need for better co-design of algorithms, numerics and computing platforms to address the challenges of workloads often run in data centers.

cs.LG↗

arXiv2018

Employing p-CSMA on a LoRa Network Simulator

Low-Power Wide-Area Networks (LPWANs) emerged to cover the needs of Internet of Things (IoT)-devices for operational longevity and long operating range. Among LPWANs, Long Range (LoRa) WAN has been the most promising; an upcoming IoT protocol, already adopted by big mobile operators like KPN and TTN. With LoRaWANs, IoT-devices transmit data to their corresponding gateways over many kilometers in a single hop and with 1% duty-cycle. However, in a LoRa network, any device claims the channel for data-transmission without performing channel-sensing or synchronization with other devices. This increases humongously the number of collisions of information-packets when the number of IoT-devices that are connected per gateway increases. To improve the utilization of the channel, we propose the application of persistent-Carrier Sense Multiple Access (p-CSMA) protocols on the MAC layer of LoRaWANs. In this manuscript, we report on the initial design of a p-CSMA component for the simulation of LoRa networks in ns3. In particular, the classes adding p-CSMA functionality to the IoT-devices are presented. Additionally, the dependencies and relations between these classes and an existing LoRaWAN module on which they apply are detailed. Further, we evaluate this new p-CSMA LoRaWAN module in terms of Packet Reception Ratio (PRR) by simulating LoRa networks. The current report is the first step in the creation of a holistic p-CSMA module, directed to support network-researchers and connoisseurs in simulating all aspects of LoRa networks in ns3.

cs.NI↗