Aug23 [Slides]: HPC on Azure in the Financial Services Industry by Darko Mocelj

In this session, Darko will provide us with an overview about HPC on the Azure Cloud and provide some insights into how HPC is used by customers in the financial services industry today.

About Darko:

As part of the EMEA Global Blackbelt (GBB) Team, I will continue supporting digital transformation in Financial Services and Life Sciences. Digital Transformation can occur on various levels across the organisation. In this role I will focus on supporting cloud transformation of advanced high-performance compute workloads for both - classical compute- and ML/AI driven workloads.

My background as a technical consultant, paired with experience working in the Financial Services sector, gives me the unique opportunity to bridge the (still separate) worlds of technology and business, paired with the possibility and work with customers across EMEA on latest technology developments.

LinkedIn: https://www.linkedin.com/in/darko-mocelj/

Transcript

1. Classified as Microsoft Confidential
   Zürich Meetup
   Azure High Performance Computing
   Dr. Darko Mocelj
   Simplify and optimize HPC deployments with Microsoft Azure
   

   View Slide

2. © Copyright Microsoft Corporation. All rights reserved.
   Classified as Microsoft Confidential
   HPC in a nutshell
   

   View Slide

3. Classified as Microsoft Confidential
   © Microsoft Corporation
   High Performance Computing (HPC) is a platform for
   groundbreaking scientific discoveries and game-changing innovation.
   Why HPC matters
   Drives innovation across
   nearly every industry to
   solve scientific and real-life
   problems such as fluid
   dynamics, finite element
   analysis, weather modelling,
   gene research, financial risk
   and more
   Furnishes fast
   computation power and
   large-scale parallel
   processing for accurately
   processing large volumes
   of data
   Provides deep insights
   into business data for
   driving smarter simulations
   and empowering
   intelligent decision making
   Saves time and money
   by delivering faster results
   

   View Slide

4. Classified as Microsoft Confidential
   Materials
   Science
   Clinical Trial
   Simulation
   Rocket
   Design
   Self-driving
   Cars
   Seismic and
   Reservoir
   modelling
   Materials
   Science
   Fundamental
   Science
   Architecture
   &
   Engineering
   Entertainment
   Genomics
   Drug Design
   Environment
   Impact
   Computational
   Chemistry
   Quantum
   computing
   Cancer
   Research
   Circuit
   Design
   Risk
   Management
   Diagnostics
   Crash Testing
   Machine Learning
   Product Design &
   Safety Security/
   Encryption
   Logistics
   Data
   Science
   

   View Slide

5. Classified as Microsoft Confidential
   HPC
   Automotive
   Oil & Gas
   Ship engineering
   Banking
   Insurance
   Energy
   Defense &
   Aerospace
   Pharmaceutical
   Healthcare
   Life Science
   Weather forecasting
   Chemical
   engineering
   Engineering &
   construction
   Graphics &
   rendering
   Fluid
   dynamics
   Structural
   simulations
   Crash
   Simulations
   Deep
   Learning
   Genomics
   Molecular
   Modelling
   Risk
   Simulations
   High Performance Computing Solutions
   Electronic
   Design
   

   View Slide

6. © Copyright Microsoft Corporation. All rights reserved.
   Classified as Microsoft Confidential
   Solutions for various workloads
   Loosely coupled
   Large-scale, compute-intensive
   workloads, which can be run in
   parallel, taking advantage of the
   scale and flexibility of the cloud
   (FSI, Genomics, Physics…)
   Tightly coupled
   Solving the underlying
   mathematical model of a dynamic
   physical system in a highly iterative
   and closely coupled fashion
   (Seismic/Reservoir, Engineering,
   Weather,…..)
   Hybrid and cloud bursting
   Optimizing application
   workflows to benefit from
   both on- and off-premises
   resources
   (FSI, Engineering, Auto, …)
   

   View Slide

7. © Copyright Microsoft Corporation. All rights reserved.
   Classified as Microsoft Confidential
   ©Microsoft Corporation
   General-purpose VMs
   D: Standard workloads
   E: High memory
   F: Compute-bound
   Solve any HPC, AI workload—at any scale
   Small scale MPI
   (Handful of cores)
   Extreme scale MPI
   (100k+ cores)
   High memory VMs
   L: High SSD & IOPS
   M: Extreme memory
   A/B series VMs
   Burstable virtual
   machines (VMs)
   D/E/F
   Cray in Azure
   Managed custom
   bare-metal server
   Large to extreme-scale
   HPC infrastructure
   Azure network integration
   Specialized VMs
   H: High memory
   HB: Memory bandwidth
   HC: Compute-bound
   NC: GP-GPU compute
   ND: Deep learning
   NV: Graphics applications
   NP: Programmable FPGA
   L/M
   H/N
   A/B
   Infiniband network interconnected
   (up to 400GB/s)
   

   View Slide

8. © Copyright Microsoft Corporation. All rights reserved.
   Classified as Microsoft Confidential
   Purpose-built HPC A full range of CPU and GPU capabilities
   that help applications scale to 80K+ cores
   Fast, secure
   networking
   Fast InfiniBand inter-connects as well
   as edge-to-cloud connectivity​
   High performing
   storage
   A range of storage capabilities to support
   simple-to-complex storage needs
   Workload
   orchestration
   End-to-end workflow agility using known,
   familiar tools and processes
   Intelligence
   services
   AI, machine learning, and deep learning
   at supercomputer scale​
   Achieve more with Microsoft Azure HPC
   

   View Slide

9. © Copyright Microsoft Corporation. All rights reserved.
   Classified as Microsoft Confidential
   HPC FinOps – FSI Focus
   

   View Slide

10. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    Fixed cost
    (on-premises)
    Variable cost
    (cloud)
    Reduce infrastructure costs
    and use capacity on demand
    Respond in an agile way to
    new business demands
    Improve operations
    Manage bursts due to
    external events quickly and
    seamlessly
    Data processing demand
    Azure HPC Cost Optimization Strategies
    Mix &
    Match
    

    View Slide

11. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    Azure HPC Cost Optimization Strategies
    Mix &
    Match
    Leverage autoscaling to
    adjust consumed compute
    capacity
    Average e uest E ecution time
    um er o e uests
    s s s s seconds minutes hours
    Understand utilizations
    patterns (refer to Azure
    Advisor) and cover the base
    utilization with Reserved
    Instances
    Compute instances can be evicted
    any time – engineer for flexibility
    and portability.
    Flexibility in selecting different
    VM families
    HBv3 Dasv5 Easv3
    F- D- E-Series
    Portability* to
    different Azure
    Regions with Spot
    instances
    availability
    

    View Slide

12. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    Azure Cost Optimization Strategies (FSI Focus)
    Workload
    Examples
    • Predictable demand, e.g., opening
    of new trading books
    • Increase Accuracy in risk models
    (replace approximations, increase
    Monte-Carlo paths, etc.)
    • Increased computational demand
    imposed by regulations (FRTB, XVA)
    Spot:
    • Grid workloads with mature
    cloud customers leveraging
    regional and compute flexibility
    • Dev / Test workloads
    • One-Off Regression Test
    • Try-Out of new architectures
    (GPU / CPU)
    • Full Flexibility
    • React quickly to volatile markets
    and adjust computational capacity
    accordingly
    

    View Slide

13. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    HPC – Know your environmental impact
    

    View Slide

14. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    Carbon-aware computing – impact on HPC operations
    https://news.microsoft.com/de-ch/2023/01/10/carbon-aware-computing-whitepaper/
    Measure carbon intensity of
    - past workloads
    - Predict future workloads and
    Minimize carbon footprint of your workloads
    Measuring the Carbon Intensity of AI in Cloud Instances (arxiv.org) https://github.com/Green-Software-Foundation/carbon-aware-sdk
    

    View Slide

15. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    Azure HPC Services
    

    View Slide

16. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    © Microsoft Corporation
    Azure Batch
    • HPC-as-a-Service Model
    • All HPC resources are cloud-based
    Services for HPC Workload Management – Azure Batch focus of the Demo
    Cloud-native job scheduling
    • Support for third party schedulers
    • Traditional HPC scaling methodology, but using Azure
    Azure CycleCloud
    Traditional cluster scheduler orchestration
    HPC App, Head node
    and on prem compute
    VM
    Resource
    Pool
    Azure
    CycleCloud
    HPC Application on
    client workstation
    VM Resource
    Pool
    Azure
    CycleCloud
    Head
    Node
    Hybrid / cloud bursting model Cloud native model
    HPC Application on
    client workstation
    VM Resource
    Pool
    Azure
    Batch
    

    View Slide

17. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    Developer access
    SaaS platform for cloud-
    enabling workflows
    .NET, Java, Node.js, Python,
    REST
    + Common languages and
    frameworks
    Job scheduling
    Easiest way to run batch
    jobs at scale in Azure
    Detect and retry failed tasks
    Task dependencies
    Job prep and cleanup tasks
    Monitoring
    VM monitoring and auto-
    recover
    Metrics and logs available via
    Portal and API
    Azure Batch Capabilities
    Autoscale
    Native VM orchestration
    Automatically scale
    environments up as down as
    jobs require
    Choice of VMs
    Windows or Linux
    Standard or custom images
    Windows pool can use AHUB
    Can use low-priority & Spot VMs
    

    View Slide

18. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    © Microsoft Corporation
    • Configure and create VMs to cater for any scale: tens to
    thousands
    • Automatically scale the number of VMs to maximize utilization
    • Easy low-priority and VM sizing, suited to your application
    Batch jobs and tasks
    • Task is a unit of execution; task = application command line
    (EXE, BAT, CMD, PS1, etc.)
    • Jobs are created and tasks are submitted to a pool. Next, tasks
    are queued and assigned to VMs
    • Any application, any execution time; run applications
    unchanged
    • Automatic detection and retry of frozen or failing tasks
    Batch pools
    Azure Batch (example of a cloud native HPC service)
    

    View Slide

19. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    Example of a Azure HPC Deployment (following Microsoft Cloud Adoption Framework for Azure)
    HPC
    Microsoft Cloud Adoption Framework for Azure - Cloud Adoption Framework | Microsoft Learn What is an Azure landing zone? - Cloud Adoption Framework | Microsoft Learn
    Connectivity /
    Shared Services
    HPC Workload
    Container
    Registry Subnet
    Container
    Registry
    

    View Slide

20. © Copyright Microsoft Corporation. All rights reserved.
    Classified as Microsoft Confidential
    You have questions? Let’s stay in touch.
    

    View Slide