Upload
others
View
5
Download
0
Embed Size (px)
Citation preview
Solution briefCisco public
Digital transformation and the emergence of AI/ML workloads have created a new set of challenges for line-of-business owners to data engineers and IT staff. Data has become an organization’s most important asset, and, with the right strategy and system, AI/ML has created an unprecedented opportunity to:
• Monetize new business models • Glean predictive insights • Enable competitive differentiation• Improve business processes
The compute layer is transforming with the rise of GPU computing enabling massive parallelism and multiple petaflops of computational power. Storage needs to be designed to provide the same degree of scale and parallelism.
Additionally, AI/ML workflows have distinct storage I/O requirements. Stages exist within data pipelines that require massive ingest and training bandwidth, mixed read/write handling, metadata labeling, and lifecycle management. SwiftStack has unique, differentiated capabilities to meet these requirements. The SwiftStack and Cisco partnership brings together best-of-class products, making it easy to consume a complete AI/ML solution stack, so organizations can focus on deriving business value.
Accelerate Deep Learning with an Edge to Core to Cloud Data Management Solution Cisco and SwiftStack collaborate for distributed, large-scale AI/ML data pipelines
Highlights• Accelerate and scale your data
pipelines with 100 GB/s+ throughput at 15 PB+ scale
• Leverage SwiftStack 1space to bring cloud and AI (AWS, GCP) to your data, and build multicloud workflows across edge, core, and cloud
• Use rich metadata tagging and search provided by SwiftStack middleware and SwiftStack Client
• Access the same namespace with both file protocols and object APIs
• Depend on a best-of-class solution stack
Solution briefCisco public
AI/ML/DL storage workflow phases and their I/O challenges
> ETL, metadata labeling> Build models - mixed R/W
Massive concurrencyand write throughput
Massive concurrency andread throughput, scale
At the edge or core, lowlatency, high throughput
Lifecycle management,governance High throughput and
concurrency tomatch edge ingestand GPU compute
Multicloud AI/ML pipelines
Policy-driven metadatalabeling, search
Challenges atthe core
Retain
Infer Train
Enrich
Ingest1
2
34
5
Use cases• Autonomous vehicles• Healthcare and life sciences• Retail and smart cities
Artificial intelligence in the data center at scaleCisco® machine-learning computing solutions ease the challenges faced by IT organizations and data scientists: supporting the needs of Machine-Learning (ML) workloads while making them part of the enterprise data center. With Cisco solutions, you can power Artificial Intelligence (AI) workloads at scale and help extract more intelligence from data to make better decisions.
SwiftStack has collaborated with Cisco to put together a reference design with Cisco UCS® S3260 Storage Servers, Cisco Nexus® 3232C 100-GbE network switches, and Cisco UCS
C480 ML M5 GPU dense servers for compute. This on-premises stack deploys a variety of AI/ML frameworks from the NVIDIA GPU Cloud (NGC) repository of easy-to-consume containers, designed for Kubernetes orchestration. In addition, this environment can be extended seamlessly with SwiftStack 1space to AWS or GCP for cloud-bursting and multicloud workflows. Businesses can quickly start their AI/ML journey, expand to leverage multicloud workflows and frameworks, and scale to petabytes of storage and 100 GBs of throughput to meet their requirements.
Solution briefCisco public
SwiftStack Cisco | AI reference architecture
Copyright © 2019 SwiftStack Inc. | swiftstack.com
© 2019 Cisco and/or its affiliates. All rights reserved. Cisco and the Cisco logo are trademarks or registered trademarks of Cisco and/or its affiliates in the U.S. and other countries. To view a list of Cisco trademarks, go to this URL: https://www.cisco.com/go/trademarks. Third-party trademarks mentioned are the property of their respective owners. The use of the word partner does not imply a partnership relationship between Cisco and any other company. (1110R) C22-742461-00 06/19
To try SwiftStack for free, go to http://www.swiftstack.com/try-it-now/.
For additional assistance or to learn more, always feel free to contact us. We’re here to help.
Phone - (415) 625-0293
Email - [email protected]
Chat - Just go to swiftstack.com and look for the chat pop-up in the bottom right.
Storage Cisco UCS S3260 Storage Servers
GPU Compute Cisco UCS C480 ML M5 Rack Server
Networking Cisco Nexus 3232C 100-GbE Switches
AI/ML FrameworksOrchestrate across public clouds and on-premises data centers
Reference architecture withSwiftStack and Cisco
AWS GoogleCloud
NvidiaGPU Cloud
TensorFlow KubeFlow PyTorch Caffe2 Valohai
AI/ML frameworks • On-premises, container-based NVIDIA GPU
Cloud (NGC) deep learning stack, with the capacity to extend to AWS and GCP
• Kubeflow for deep learning workflow management
GPU compute• The Cisco UCS C480 ML M5 Rack Server
is designed for the most compute-intensive phase of the AI and ML lifecycle: deep learning
• This server integrates GPUs and high-speed interconnect technology combined with large storage capacity and up to 100-Gbps network connectivity
Networking• Two Cisco Nexus 3232C 100-GbE switches
Storage• SwiftStack software for scale-out multicloud
storage and data management• Cisco UCS S3260 Storage Servers with 1.5PB
of capacity. Erasure coding or replication-based data protection policies
• Scale: Increase Cisco UCS C480 ML M5s for additional PFLOPS and Cisco UCS S3260s for PBs of storage and GBs of throughput