• Summary
• Contents:

EBS - General Purpose SSD

Ref: https://learn.cantrill.io/courses/1820301/lectures/41301321

• 🔧 Balance of price and performance
  • Flexible type of storage → good for general usage
  • 💡 Was game breaker when introduced → high-performing storage for a fairly low price
• Size range: 1GB-16TB
• ‼️ Performance standardization: 1 IO = 1 (IO) credit = 16kB ! (for all SSD volumes)
  • This is how AWS defines and measures IOPS for billing and performance in EBS SSD volumes
  • 1 IOPS = 1 IO in 1 sec = 1 credit/s = 16kB/s
• Two types:
  1. GP2 (default)
  2. GP3 (new, generally cheaper, probably will become default)
• Recommended for most workloads:
  • Boot volumes
  • Low-latency interactive apps
  • Dev & Test environments
  • GP3 also good for:
    • Virtual desktops
    • Medium sized single instance DBs (e.g. MSSQL Server, Oracle DB…)
• 💡 Sadly, you have to commit the size and min-max performance numbers for GP2 and GP3 volumes to memory in order to exam a few exam questions 😞. They help you discern between correct and incorrect options.

EBS GP2 Volumes

Diagram: https://github.com/acantril/aws-sa-associate-saac03/blob/main/0900-ELASTIC_COMPUTE_CLOUD(EC2)/00_LEARNINGAIDS/EBSVolumes-1.png

• Default general purpose SSD storage provided by EBS
• Performance
  • Baseline performance rate
    • 🔧 Determines the minimum (baseline) IOPS that can be performed on disk
    • Depends on disk size
      • 1 GB ≤ Size ≤ 33.33GB → Baseline = 100 IOPS (GP2 minimum baseline)
      • 33.33 GB ≤ Size ≤ 5.33TB → Baseline = 3 IOPS per GB
        • e.g. Baseline[100GB] = 3x100 = 300 IOPS
      • 5.33 TB ≤ Size ≤ 16TB → Baseline = 16 000 IOPS (GP2 maximum baseline)
        • GP2 maximum throughput = 16kB (one IO credit) * 16 000 IOPS ≈ 250 MB/s
  • Burst performance rate = 3000 IOPS
    • Only relevant for volumes < 1TB → Determines the maximum IOPS that can be performed on disk
    • Volumes > 1TB have a baseline of more than 3000 IOPS → burst rate irrelevant
• Volumes < 1TB follow an IO credit allocation model
  • Conceptually: credit bucket
    • Credits spent in disk operations
    • Bucket capacity = 5.4 million credits (starts full)
    • Bucket fills at baseline performance rate of volume
  • Performance scenarios:
    1. Actual Rate < Baseline Rate → Bucket refills
    2. Actual Rate = Baseline Rate → Bucket quantity stays the same
       • ❗ Still fine even if bucket is empty! (since bucket is being refilled at same rate)
    3. Baseline Rate < Actual Rate ≤ Burst Rate → Bucket is depleted
       • ‼️ If bucket is empty → Cost penalties!!! (avoid!)
  • Credit bucket starts up full with 5.4 million credits
    • Allows 30 mins at 3000 IOPS (burst rate) assuming no refilling
      • 👍 Great for booting and initial workloads
      • Could burst much longer than 30 mins because bucket is always refilling
  • 💡 Key requirement: manage all credit buckets of all of your volumes (ensure they stay replenished and that you're not penalized)
• ❗ Volumes ≥ 1TB don't use the credit model
  • Baseline rate exceeds burst rate → Volumes always perform with baseline rate

EBS GP3 Volumes

Diagram: https://github.com/acantril/aws-sa-associate-saac03/blob/main/0900-ELASTIC_COMPUTE_CLOUD(EC2)/00_LEARNINGAIDS/EBSVolumes-2.png

• Newer type of storage, will probably become the default over GP2 at some point
  • More performant, and generally cheaper than GP2 volumes!