Recap

• Week 5 Feedback Form: ter.ps/389L5
• Codelab 3 (EC2) Released: ter.ps/389LCL3
  • No submission necessary
  • However, make sure that you understand the concepts

New Student Facilitator

• Andrej Rasevic will be joining as a second student facilitator.
• He will be helping out with codelabs + projects

• Contact: arasevic@cs.umd.edu

EC2

EC2 Pricing

• On-Demand Pricing
  • t2.micro: ​$0.0116 per hour (pro-rated to the second!)
    • $8.352 / month, if run 24/7
• Free Tier:
  • 750 hours / month (of t2.micro)
    • 744 hours in a 31-day month

EC2 Pricing

• On-Demand Pricing (cont.)
  • smallest: t2.nano for $4.176 / month
  • largest: x1e.32xlarge (4TB memory): $26.688 / hour
    • ~$20k/month
• Other options:
  • Spot Pricing: Bid on instance pricing in a market
    • AWS's spare compute power
    • Can be up to 90% cheaper
  • Reserved Pricing: Request usage in advance and in bulk
    • Good for predictable usage
    • Can be up to 75% cheaper

EC2 Pricing

• Data Transfer
  • ​In to / Out from AWS:
    • Same AZ: Free
    • Different AZ/Region*: 1-2¢ / GB
  • Out to internet:
    • 9¢ / GB
• * = free for some AWS services in the same region (f.e. S3, DynamoDB, SQS)

EBS Pricing

• General Purpose SSD Volumes: 10¢ / GB
  • ​40GB EBS volume used for one month is $4

SQS

Message Queues

• Message Queue: A shared queue that enables asynchronous communication of messages between processes.

SQS

• Simple Queue Service (SQS): A managed, distributed message queueing service frequently used to communicate between AWS services.

Why decouple components?

• Strongly coupled components form complex systems
  • Hard to maintain/expand/refactor
• Instead:
  • Encapsulate the inner workings of each component
  • Expose a limited API

Why decouple components?

Why use message queues?

• What happens if the Service B dies?
• What happens if Service B can't process requests as quickly as Service A produces them?
  • Load spikes?
• What if Service B dies while processing a request?

Real World Use Case

• Nextdoor sends ~30M emails / day
  • ~350/second, but very spiky
• 100s of components that need to send emails
• Single service handles email generation and sending
• Components simply call an API from a library which places a message into an SQS queue

SQS Concepts

• Delivery:
  • ​Want to guarantee that all messages are processed​

• We can only guarantee At-Lease-Once Delivery
  • ​Every message will be delivered
  • But, it may be delivered more than once

SQS Concepts

• Ordering:
  • ​We want high throughput of messages

• Standard Queues: "Unlimited" TPS
• FIFO Queues: Max 300 TPS

SQS Concepts

• Polling:
  • How do receivers retrieve messages from the queue?
    • Long Polling: Wait until a message is available for processing. Searches all nodes.
    • Short Polling: Immediately return if no messages available. Search subset of nodes.

SQS Concepts

• Visibility Timeouts:
  • When a receiver receives a message, that message is hidden from other receivers
  • After a predefined amount of time (a few seconds up to 12 hours), it is made visible again

SQS Concepts

• Retention Period: Messages are deleted from a queue if not processed within a certain amount of time (up to 14 days)
• Max Message Size: Up to 256KB per message
• Dead Letter Queue: Messages can be placed in a DLQ after failing to be deleted after a predefined number of receives

SQS Demo

SQS Pricing

• Requests
  • Standard: 40¢ per 1M requests
  • FIFO: 50¢ per 1M requests
• Data Transfer between SQS/EC2
  • Free (if within the same region)

• Can return multiple messages per request
• 1 request = 64KB chunk of a payload

• Free Tier: 1M requests

SQS Worksheet

Complete the worksheet section on SQS.