Extra Solution Architecture and AWS Services discussions

Lambda, SNS & SQS

• SQS + Lambda
• SQS FIFO + Lambda
• SNS asynchroous to Lambda + SQS

  Fan Out Pattern: deliver to multiple SQS

• SDK put SQSs
• SDK put SNS subcrible SQS(fan out)

  S3 Events

• S3: objectCreated, S3:ObjectRemove, S3:ObjectRestore, S3:Replication…
• Object name filtering possible (*.jpg)
• Use case: generate thumbnails of images upload to S3
• Can create as many “S3 events” as desired
• S3 event notificatiions typically deliver events in secconds butt can sometimes take a minutesor longer
• If two writes are made to a single non-versioned object at the same time, it is possible that only a single event notification will be sent
• If you want to ensure tthat an event notification is sent for every successful write you can enable versioning on your bucket

  Caching Strategies

• Clientt -> CloudFront -> API gateway -> App Logic Ec2 / Lambda -> Redis Memcached -> Database
• Cliennt -> CloudFront(Edge) -> S3
• Caching, TTL, Network, Computation, Cost, Latency

  Blocking an IP address

• EC2 Instance Public IP
• Optional Firewall Software in EC2s
• NACL + Security groups

  Blocking an IP address – with an ALB

• Application Load Balancer Connection Termination
• EC2 Instance Private IP

  Blocking an IP address – with an NLB

• Network Load Balancer Traffic goes through No Security Group

  Blocking an IP address – ALB + WAF

• ALB + WAF(IP address filtering)

  Blocking an IP address – ALB, CloudFront WAF

• Client -> CloudFront(GEO Restriction) -> WAF IP address filtering -> CloudFrontt Public IPs -> Public ALB -> EC2 Security group

  High Performance Computing (HPC)

• The cloud is perfect place to perform HPCC
• You can create a very high number of resources in no time
• You can speed up time to results by adding more resources
• You can pay only for the systems you have used
• Perform genomics, computation chemistry, financial risk modeling, weather predictionn, machine learning, deep learning, autonomous driving
• Which services help perform HPC?

  Data Management & Transfer

• AWS Direct Connect
  • Move GB/s of data to the cloud, over a private secure network
• Snowball & Snowmobile
  • Move PB of data to the cloud
• AWS DataSync
  • Move large amount of datta between on-premises and S3, EFS, FSc for Windows

    Compute and Nettworking

• EC2 Instances:
  • CPU optimized, GPU opttimized
  • Spot Instances / Spot Fleets for cost savings + Auto Scaling
• EC2 Placement Groups: Cluster for good network performance
• Placement group cluster low latency 10Gbps network

  Compute and Networking

• EC2 Enhaced Networking (SR-IOV)
  • Higher bandwidth, higher PPS (packet per second), lower latency
  • Option1: Elastic Network Adapter (ENA) up to 100 Gbps
  • Option2: Intel 82599 VF up to 10 Gbps - LEGACY
• Elastic Fabric Adapter (EFA)
  • Improved ENA for HPC, only works for linux
  • Greate for inter-node communication, tighly coupled workloads
  • Leverages Message Passing Interface (MPI) standard
  • Bypasses the underlying Linux OS to provide low-latenccy, reliable trannsport

    Storage

• Instance-attached storage
  • EBS: scale up to 256000 IOPS with io2 Block Express
  • Instance Store: Scale tto millions of IOPS, linked to EC2 instancee, low latenccy
• Network storage:
  • Amazon S3: large blob, not a file system
  • Amazon EFS: scale IOPS based on ttottal size, or use provisioned IOPS
  • Amazon FSx for Lustre:
    • HPC optimized distributed file system, milions of IOPS
    • Backed by S3

      Automation and Orchestration

• AWS Batch
  • AWS BAtch supports multi-node parallel jobs, which enables you to run single jobs that span multtiple EC2 instances
  • Easily schedule jobs and launch EC2 instances accordingly
• AWS ParallelCluster
  • Open-source cluster management tool to deploy HPC on AWS
  • Configure with text files
  • Automate creation of VPC, subnet, clusterr type and instancce types
  • Abiliitty to enable EFA on the cluster (improves nnetwork performance)

    Creating a highly available EC2 instance

    Creating a highly available EC2 instance With an Auto Scaling Group

    Creating a highly available EC2 instance width ASG + EBS

    Creating a highly available EC2 instance

• HA options for the bastion host
  • Run2across2AZ
  • Run 1 across 2 AZ with 1 ASG 1:1:1
• Routing to the bastion host
  • If 1 bastion host, use an elastic IP with
  • ec2 user-data script to access it
  • If 2 bastion hosts, use an Network Load Balancer (layer 4) deployed in multiple AZ
  • If NLB, the bastion hosts can live in the private subnet directly
• Note: Can’t use ALB as ALB is layer 7(HTTP protocol)