Great question—this is one of the most common gotchas with AWS Lambda that trips up a lot of developers! Let me break down exactly what's happening here:

• Lambda's resource model is memory-bound, not just memory-focused
  When you pick a memory size for your Lambda function, you're not just setting a limit on how much RAM it can use. AWS ties CPU power, network bandwidth, and I/O throughput directly to the memory allocation. For example:

  • A 1024MB Lambda gets the full baseline CPU capacity (roughly equivalent to a dedicated core for practical purposes).
  • A 128MB Lambda only gets ~1/8 of that CPU capacity.
    So even if your code only needs 22MB of RAM, the CPU it's running on is drastically slower when you drop to 128MB. Any CPU-bound work in your function—like JSON serialization/deserialization, data parsing, encryption, or even lightweight processing loops—will take longer to complete with the reduced CPU power.

• Network performance scales with memory too
  If your function makes any network calls (to APIs, S3, DynamoDB, etc.), the available bandwidth also decreases as you lower the memory size. A 128MB Lambda has significantly less network throughput than a 1024MB one, so even a simple HTTP request might take longer to send/receive data, adding to your total execution time.

• You're not seeing the full picture in memory usage
  The "Max Memory Used" metric only tracks the RAM your application code consumes, but Lambda runs your code in a runtime environment (like Node.js, Python, etc.) that has its own overhead. Even so, that overhead is unlikely to push you near 128MB—your core issue is still the tied CPU/network resources.

Quick cost vs performance check #

Just to put this in perspective: Lambda billing is calculated as (memory size / 1024) * duration in seconds. Let's compare your two scenarios:

• 1024MB * 0.12s = 0.12 GB-seconds
• 128MB * 0.35s = 0.04375 GB-seconds

So even though the 128MB function is slower, it's cheaper. But if latency matters for your use case, you'll want to pick a memory size that gives you enough CPU/network performance to keep execution time low—even if your code doesn't use all the RAM.

内容的提问来源于stack exchange，提问作者sigmaxf