LogoLogo
Continuum WebsiteContinuum ApplicationsContinuum KnowledgeAxolotl Platform
Continuum Knowledge
Continuum Knowledge
  • Continuum
  • Data
    • Datasets
      • Pre Training Data
      • Types of Fine Tuning
      • Self Instruct Paper
      • Self-Alignment with Instruction Backtranslation
      • Systematic Evaluation of Instruction-Tuned Large Language Models on Open Datasets
      • Instruction Tuning
      • Instruction Fine Tuning - Alpagasus
      • Less is More For Alignment
      • Enhanced Supervised Fine Tuning
      • Visualising Data using t-SNE
      • UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
      • Training and Evaluation Datasets
      • What is perplexity?
  • MODELS
    • Foundation Models
      • The leaderboard
      • Foundation Models
      • LLama 2 - Analysis
      • Analysis of Llama 3
      • Llama 3.1 series
      • Google Gemini 1.5
      • Platypus: Quick, Cheap, and Powerful Refinement of LLMs
      • Mixtral of Experts
      • Mixture-of-Agents (MoA)
      • Phi 1.5
        • Refining the Art of AI Training: A Deep Dive into Phi 1.5's Innovative Approach
      • Phi 2.0
      • Phi-3 Technical Report
  • Training
    • The Fine Tuning Process
      • Why fine tune?
        • Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations?
        • Explanations in Fine Tuning
      • Tokenization
        • Tokenization Is More Than Compression
        • Tokenization - SentencePiece
        • Tokenization explore
        • Tokenizer Choice For LLM Training: Negligible or Crucial?
        • Getting the most out of your tokenizer for pre-training and domain adaptation
        • TokenMonster
      • Parameter Efficient Fine Tuning
        • P-Tuning
          • The Power of Scale for Parameter-Efficient Prompt Tuning
        • Prefix-Tuning: Optimizing Continuous Prompts for Generation
        • Harnessing the Power of PEFT: A Smarter Approach to Fine-tuning Pre-trained Models
        • What is Low-Rank Adaptation (LoRA) - explained by the inventor
        • Low Rank Adaptation (Lora)
        • Practical Tips for Fine-tuning LMs Using LoRA (Low-Rank Adaptation)
        • QLORA: Efficient Finetuning of Quantized LLMs
        • Bits and Bytes
        • The Magic behind Qlora
        • Practical Guide to LoRA: Tips and Tricks for Effective Model Adaptation
        • The quantization constant
        • QLORA: Efficient Finetuning of Quantized Language Models
        • QLORA and Fine-Tuning of Quantized Language Models (LMs)
        • ReLoRA: High-Rank Training Through Low-Rank Updates
        • SLoRA: Federated Parameter Efficient Fine-Tuning of Language Models
        • GaLora: Memory-Efficient LLM Training by Gradient Low-Rank Projection
      • Hyperparameters
        • Batch Size
        • Padding Tokens
        • Mixed precision training
        • FP8 Formats for Deep Learning
        • Floating Point Numbers
        • Batch Size and Model loss
        • Batch Normalisation
        • Rethinking Learning Rate Tuning in the Era of Language Models
        • Sample Packing
        • Gradient accumulation
        • A process for choosing the learning rate
        • Learning Rate Scheduler
        • Checkpoints
        • A Survey on Efficient Training of Transformers
        • Sequence Length Warmup
        • Understanding Training vs. Evaluation Data Splits
        • Cross-entropy loss
        • Weight Decay
        • Optimiser
        • Caching
      • Training Processes
        • Extending the context window
        • PyTorch Fully Sharded Data Parallel (FSDP)
        • Train Short, Test Long: Attention with Linear Biases Enables Input Length Extrapolation
        • YaRN: Efficient Context Window Extension of Large Language Models
        • Sliding Window Attention
        • LongRoPE
        • Reinforcement Learning
        • An introduction to reinforcement learning
        • Reinforcement Learning from Human Feedback (RLHF)
        • Direct Preference Optimization: Your Language Model is Secretly a Reward Model
  • INFERENCE
    • Why is inference important?
      • Grouped Query Attention
      • Key Value Cache
      • Flash Attention
      • Flash Attention 2
      • StreamingLLM
      • Paged Attention and vLLM
      • TensorRT-LLM
      • Torchscript
      • NVIDIA L40S GPU
      • Triton Inference Server - Introduction
      • Triton Inference Server
      • FiDO: Fusion-in-Decoder optimised for stronger performance and faster inference
      • Is PUE a useful measure of data centre performance?
      • SLORA
  • KNOWLEDGE
    • Vector Databases
      • A Comprehensive Survey on Vector Databases
      • Vector database management systems: Fundamental concepts, use-cases, and current challenges
      • Using the Output Embedding to Improve Language Models
      • Decoding Sentence-BERT
      • ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT
      • SimCSE: Simple Contrastive Learning of Sentence Embeddings
      • Questions Are All You Need to Train a Dense Passage Retriever
      • Improving Text Embeddings with Large Language Models
      • Massive Text Embedding Benchmark
      • RocketQAv2: A Joint Training Method for Dense Passage Retrieval and Passage Re-ranking
      • LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders
      • Embedding and Fine-Tuning in Neural Language Models
      • Embedding Model Construction
      • Demystifying Embedding Spaces using Large Language Models
      • Fine-Tuning Llama for Multi-Stage Text Retrieval
      • Large Language Model Based Text Augmentation Enhanced Personality Detection Model
      • One Embedder, Any Task: Instruction-Finetuned Text Embeddings
      • Vector Databases are not the only solution
      • Knowledge Graphs
        • Harnessing Knowledge Graphs to Elevate AI: A Technical Exploration
        • Unifying Large Language Models and Knowledge Graphs: A Roadmap
      • Approximate Nearest Neighbor (ANN)
      • High Dimensional Data
      • Principal Component Analysis (PCA)
      • Vector Similarity Search - HNSW
      • FAISS (Facebook AI Similarity Search)
      • Unsupervised Dense Retrievers
    • Retrieval Augmented Generation
      • Retrieval-Augmented Generation for Large Language Models: A Survey
      • Fine-Tuning or Retrieval?
      • Revolutionising Information Retrieval: The Power of RAG in Language Models
      • A Survey on Retrieval-Augmented Text Generation
      • REALM: Retrieval-Augmented Language Model Pre-Training
      • Retrieve Anything To Augment Large Language Models
      • Generate Rather Than Retrieve: Large Language Models Are Strong Context Generators
      • Active Retrieval Augmented Generation
      • DSPy: LM Assertions: Enhancing Language Model Pipelines with Computational Constraints
      • DSPy: Compiling Declarative Language Model Calls
      • DSPy: In-Context Learning for Extreme Multi-Label Classification
      • Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programs
      • HYDE: Revolutionising Search with Hypothetical Document Embeddings
      • Enhancing Recommender Systems with Large Language Model Reasoning Graphs
      • Retrieval Augmented Generation (RAG) versus fine tuning
      • RAFT: Adapting Language Model to Domain Specific RAG
      • Summarisation Methods and RAG
      • Lessons Learned on LLM RAG Solutions
      • Stanford: Retrieval Augmented Language Models
      • Overview of RAG Approaches with Vector Databases
      • Mastering Chunking in Retrieval-Augmented Generation (RAG) Systems
    • Semantic Routing
    • Resource Description Framework (RDF)
  • AGENTS
    • What is agency?
      • Rephrase and Respond: Let Large Language Models Ask Better Questions for Themselves
      • Types of Agents
      • The risk of AI agency
      • Understanding Personality in Large Language Models: A New Frontier in AI Psychology
      • AI Agents - Reasoning, Planning, and Tool Calling
      • Personality and Brand
      • Agent Interaction via APIs
      • Bridging Minds and Machines: The Legacy of Newell, Shaw, and Simon
      • A Survey on Language Model based Autonomous Agents
      • Large Language Models as Agents
      • AI Reasoning: A Deep Dive into Chain-of-Thought Prompting
      • Enhancing AI Reasoning with Self-Taught Reasoner (STaR)
      • Exploring the Frontier of AI: The "Tree of Thoughts" Framework
      • Toolformer: Revolutionising Language Models with API Integration - An Analysis
      • TaskMatrix.AI: Bridging Foundational AI Models with Specialised Systems for Enhanced Task Completion
      • Unleashing the Power of LLMs in API Integration: The Rise of Gorilla
      • Andrew Ng's presentation on AI agents
      • Making AI accessible with Andrej Karpathy and Stephanie Zhan
  • Regulation and Ethics
    • Regulation and Ethics
      • Privacy
      • Detecting AI Generated content
      • Navigating the IP Maze in AI: The Convergence of Blockchain, Web 3.0, and LLMs
      • Adverse Reactions to generative AI
      • Navigating the Ethical Minefield: The Challenge of Security in Large Language Models
      • Navigating the Uncharted Waters: The Risks of Autonomous AI in Military Decision-Making
  • DISRUPTION
    • Data Architecture
      • What is a data pipeline?
      • What is Reverse ETL?
      • Unstructured Data and Generatve AI
      • Resource Description Framework (RDF)
      • Integrating generative AI with the Semantic Web
    • Search
      • BM25 - Search Engine Ranking Function
      • BERT as a reranking engine
      • BERT and Google
      • Generative Engine Optimisation (GEO)
      • Billion-scale similarity search with GPUs
      • FOLLOWIR: Evaluating and Teaching Information Retrieval Models to Follow Instructions
      • Neural Collaborative Filtering
      • Federated Neural Collaborative Filtering
      • Latent Space versus Embedding Space
      • Improving Text Embeddings with Large Language Models
    • Recommendation Engines
      • On Interpretation and Measurement of Soft Attributes for Recommendation
      • A Survey on Large Language Models for Recommendation
      • Model driven recommendation systems
      • Recommender AI Agent: Integrating Large Language Models for Interactive Recommendations
      • Foundation Models for Recommender Systems
      • Exploring the Impact of Large Language Models on Recommender Systems: An Extensive Review
      • AI driven recommendations - harming autonomy?
    • Logging
      • A Taxonomy of Anomalies in Log Data
      • Deeplog
      • LogBERT: Log Anomaly Detection via BERT
      • Experience Report: Deep Learning-based System Log Analysis for Anomaly Detection
      • Log-based Anomaly Detection with Deep Learning: How Far Are We?
      • Deep Learning for Anomaly Detection in Log Data: A Survey
      • LogGPT
      • Adaptive Semantic Gate Networks (ASGNet) for log-based anomaly diagnosis
  • Infrastructure
    • The modern data centre
      • Enhancing Data Centre Efficiency: Strategies to Improve PUE
      • TCO of NVIDIA GPUs and falling barriers to entry
      • Maximising GPU Utilisation with Kubernetes and NVIDIA GPU Operator
      • Data Centres
      • Liquid Cooling
    • Servers and Chips
      • The NVIDIA H100 GPU
      • NVIDIA H100 NVL
      • Lambda Hyperplane 8-H100
      • NVIDIA DGX Servers
      • NVIDIA DGX-2
      • NVIDIA DGX H-100 System
      • NVLink Switch
      • Tensor Cores
      • NVIDIA Grace Hopper Superchip
      • NVIDIA Grace CPU Superchip
      • NVIDIA GB200 NVL72
      • Hopper versus Blackwell
      • HGX: High-Performance GPU Platforms
      • ARM Chips
      • ARM versus x86
      • RISC versus CISC
      • Introduction to RISC-V
    • Networking and Connectivity
      • Infiniband versus Ethernet
      • NVIDIA Quantum InfiniBand
      • PCIe (Peripheral Component Interconnect Express)
      • NVIDIA ConnectX InfiniBand adapters
      • NVMe (Non-Volatile Memory Express)
      • NVMe over Fabrics (NVMe-oF)
      • NVIDIA Spectrum-X
      • NVIDIA GPUDirect
      • Evaluating Modern GPU Interconnect
      • Scalable Hierarchical Aggregation and Reduction Protocol (SHARP)
      • Next-generation networking in AI environments
      • NVIDIA Collective Communications Library (NCCL)
    • Data and Memory
      • NVIDIA BlueField Data Processing Units (DPUs)
      • Remote Direct Memory Access (RDMA)
      • High Bandwidth Memory (HBM3)
      • Flash Memory
      • Model Requirements
      • Calculating GPU memory for serving LLMs
      • Transformer training costs
      • GPU Performance Optimisation
    • Libraries and Complements
      • NVIDIA Base Command
      • NVIDIA AI Enterprise
      • CUDA - NVIDIA GTC 2024 presentation
      • RAPIDs
      • RAFT
    • Vast Data Platform
      • Vast Datastore
      • Vast Database
      • Vast Data Engine
      • DASE (Disaggregated and Shared Everything)
      • Dremio and VAST Data
    • Storage
      • WEKA: A High-Performance Storage Solution for AI Workloads
      • Introduction to NVIDIA GPUDirect Storage (GDS)
        • GDS cuFile API
      • NVIDIA Magnum IO GPUDirect Storage (GDS)
      • Vectors in Memory
Powered by GitBook
LogoLogo

Continuum - Accelerated Artificial Intelligence

  • Continuum Website
  • Axolotl Platform

Copyright Continuum Labs - 2023

On this page
  • Introduction to WEKA
  • Key Characteristics of WEKA
  • Why WEKA's Architecture Excels for AI Workloads

Was this helpful?

  1. Infrastructure
  2. Storage

WEKA: A High-Performance Storage Solution for AI Workloads

PreviousStorageNextIntroduction to NVIDIA GPUDirect Storage (GDS)

Last updated 10 months ago

Was this helpful?

Introduction to WEKA

WEKA is a software-defined storage solution that provides a distributed file system called WekaFS.

It is designed to meet the demanding performance and scalability requirements of modern workloads, particularly in the fields of artificial intelligence (AI), machine learning (ML), high-performance computing (HPC), and financial trading.

WekaFS runs on standard x86 servers and NVMe SSDs, eliminating the need for specialised hardware. This allows organisations to leverage the latest advancements in hardware technology without being tied to proprietary storage solutions.

Key Characteristics of WEKA

  1. High Performance: WEKA leverages NVMe flash storage and a distributed architecture to provide high throughput and low latency, enabling it to handle AI and HPC workloads.

  2. Scalability: WEKA is highly scalable, both in terms of capacity and performance. It can scale to petabytes of storage and billions of files, while maintaining consistent performance.

  3. Unified Namespace: WEKA provides a single, unified namespace that spans both high-performance flash storage and cost-effective object storage, with transparent tiering between the two.

  4. Protocol Support: WekaFS supports a wide range of protocols, including POSIX, NFS, SMB, S3, and GPUDirect Storage, making it compatible with a variety of applications and workflows.

  5. Data Protection: WEKA ensures data protection through distributed erasure coding, providing resilience against multiple concurrent drive or node failures without sacrificing performance.

  6. Storage Efficiency: Features like thin provisioning, inline compression, and deduplication help optimize storage utilization and reduce costs.

  7. Ease of Management: WEKA provides a user-friendly GUI for managing the storage system, including features like snapshots, clones, and quality of service (QoS) controls.

An explanation of Weka's cloud architecture

Weka's cloud architecture is designed to provide a high-performance, scalable, and flexible storage solution for demanding workloads such as AI, machine learning, and high-performance computing. The core components and their interactions are as follows:

Weka Cluster

  • The Weka cluster consists of a group of EC2 instances (e.g., i3en instances in AWS or LSv3 instances in Azure) that run the Weka software.

  • These instances leverage their local NVMe storage to create a distributed, high-performance flash tier.

  • The Weka software aggregates and virtualizes the NVMe capacity across all instances, presenting it as a single, unified storage pool.

  • Data is distributed across the cluster using Weka's proprietary data protection scheme, which provides redundancy and fault tolerance.

Data Tiering and Object Storage Integration

  • Weka extends the storage capacity by seamlessly tiering data between the high-performance flash tier and a cheaper object storage tier (e.g., Amazon S3 or Azure Blob Storage).

  • The Weka filesystem manages the tiering process transparently, ensuring that hot data resides on the flash tier for optimal performance, while cold data is moved to the object tier for cost-effective storage.

  • Tiering granularity is at the file level, with Weka using a standard block size to chunk large files and aggregate small files for efficient tiering and metadata management.

  • Weka maintains a single namespace that spans both the flash tier and the object tier, providing a unified view of the entire dataset.

Data Access and Protocol Support

  • Applications can access the Weka filesystem using various standard protocols such as NFS, SMB, and S3, as well as Weka's native POSIX client for optimal performance.

  • In Kubernetes environments, Weka provides a CSI driver for seamless integration and dynamic provisioning of persistent volumes.

  • Weka's client software runs in user space and communicates directly with the storage nodes, bypassing the kernel to minimize latency and maximize throughput.

Deployment Automation and Scalability

  • Weka integrates with infrastructure-as-code tools like Terraform and AWS CloudFormation to automate the deployment process.

  • These tools provision the necessary cloud resources (e.g., VPC, security groups, instances) and configure the Weka cluster based on predefined templates.

  • Weka supports user-driven scaling, allowing administrators to easily add or remove storage nodes to adapt to changing capacity and performance requirements.

  • The Weka cluster automatically rebalances data across nodes when the cluster size changes, ensuring optimal data distribution and performance.

Hybrid and Multi-Cloud Capabilities

  • Weka enables hybrid and multi-cloud deployments through its "Snap-to-Object" feature, which creates a point-in-time, self-describing snapshot of the entire Weka filesystem (data and metadata) and stores it in an object store.

  • These snapshots can be replicated to object stores in different regions or cloud providers, providing a mechanism for data mobility and disaster recovery.

  • Weka clusters can be rapidly provisioned in the target cloud environment using the snapshot data, enabling use cases like cloud bursting, data migration, and multi-cloud workflows.

Converged Deployment Mode

  • In addition to the traditional dedicated storage cluster deployment, Weka supports a converged deployment mode where the Weka software runs on the same instances as the application.

  • This mode leverages the abundant compute resources (vCPUs and memory) available in modern instance types, particularly those with GPU acceleration.

  • Converged deployments can offer cost and efficiency benefits by reducing the need for separate storage infrastructure, especially in scenarios with ample instance resources.

Key Architectural Advantages

  • Weka's distributed, parallel filesystem architecture enables high performance and scalability by leveraging the aggregate bandwidth and IOPS of multiple NVMe devices.

  • The tiering mechanism extends capacity cost-effectively while maintaining a single namespace and transparent data movement between tiers.

  • Support for standard protocols and integration with Kubernetes simplifies application access and deployment.

  • Infrastructure-as-code integration and user-driven scaling streamline cluster management and enable elasticity.

  • Snap-to-Object and multi-cloud capabilities unlock data mobility, disaster recovery, and flexibility in choosing cloud providers.

  • The converged deployment mode optimizes resource utilization and reduces infrastructure complexity in suitable scenarios.

Overall, Weka's cloud architecture combines high-performance NVMe storage, intelligent tiering, and multi-cloud mobility to address the storage challenges of data-intensive workloads in the cloud.

Why WEKA's Architecture Excels for AI Workloads

Weka is designed to optimise GPU utilisation by addressing the key challenges that can lead to GPU underutilisation, such as data stalls and pipeline starvation.

Weka's architecture and features enable it to constantly feed data to GPUs, ensuring they are fully utilised.

Here's how Weka achieves this:

High-performance, low-latency storage

  • Weka is built on a distributed, parallel file system architecture that leverages NVMe flash storage.

  • By using high-speed NVMe drives and a software-defined approach, Weka provides extremely high throughput and low latency storage performance.

  • This high-performance storage ensures that data can be quickly delivered to GPUs, minimizing wait times and keeping them busy.

Efficient metadata management

  • AI and ML workloads often involve a large number of small files, which can create metadata bottlenecks.

  • Weka efficiently handles this by creating virtual metadata servers that scale dynamically with each server added to the cluster.

  • This distributed metadata architecture allows Weka to manage billions of files and trillions of metadata operations, eliminating metadata as a bottleneck.

Distributed data and metadata

  • Weka distributes and parallelises both data and metadata across the entire cluster in small, 4K chunks.

  • This distribution ensures low latency and high performance regardless of the I/O size (small, large, or mixed).

  • By spreading data and metadata across all nodes, Weka maximizes the aggregate bandwidth and IOPS of the system.

Tiered storage and data locality

  • Weka automatically tiers data between the high-performance NVMe flash tier and a cheaper object storage tier.

  • The tiering mechanism keeps hot data on the flash tier closest to the GPUs, while cold data is moved to the object tier.

  • This ensures that the most frequently accessed data is always available with the lowest latency, optimizing GPU utilization.

Direct data access and GPU-aware storage

  • Weka supports direct access to data from GPUs, bypassing the CPU and reducing latency.

  • Integration with technologies like GPUDirect Storage allows GPUs to directly read and write data to Weka storage, further reducing data movement overhead.

  • This direct data path between storage and GPU memory minimizes data stalls and keeps GPUs fed with data.

Scalability and performance scaling

  • Weka's distributed architecture allows it to scale linearly by adding more nodes to the cluster.

  • As the cluster grows, both capacity and performance increase, ensuring that storage can keep pace with the demands of additional GPUs.

  • This scalability ensures that storage performance can be easily increased to match the needs of growing GPU clusters.

Optimised data pipelines

  • Weka's architecture is designed to minimise data movement and copies throughout the AI/ML pipeline.

  • By providing a unified namespace across flash and object tiers, Weka eliminates the need for manual data copies between storage layers.

  • This streamlined data pipeline reduces latency and ensures that data is efficiently delivered to GPUs at each stage of the workflow.

WekaFS combines NVMe flash with cloud object storage in a single global namespace
Page cover image