Skip to content

instance-type-sweep-results.md

Latest commit

 

History

History
53 lines (41 loc) · 1.82 KB

instance-type-sweep-results.md

File metadata and controls

53 lines (41 loc) · 1.82 KB
Raw

Instance Type Sweep Results

Date: 2026-01-31 Region: us-west-2 Technique: Cold launch across different instance types

Summary

Instance Type API Pending→Running Boot→SSH Total
t3.medium 3718ms 16620ms 2744ms 23082ms
t3.large 3885ms 16457ms 2979ms 23321ms
m6i.large 3727ms 16553ms 2672ms 22952ms
m7i.large 3789ms 16368ms 2584ms 22741ms
c7i.large 3581ms 16368ms 2727ms 22676ms
r7i.large 3464ms 16477ms 2700ms 22641ms

Fastest: r7i.large (22641ms) Slowest: t3.large (23321ms) Variance: 680ms (2.6%)

Key Finding

Instance type has minimal impact on startup time.

The Pending→Running phase is remarkably consistent across all instance types:

  • Range: 16.4s - 16.6s
  • Variance: ~250ms

This suggests the VM startup overhead is:

  • Independent of instance family (burstable vs general vs compute vs memory)
  • Independent of instance size (at least for "large" variants)
  • A fixed cost in the AWS infrastructure layer

Implications

  1. Don't optimize instance type for speed - Pick based on workload, not startup
  2. The 16s floor is universal - Applies to all tested instance families
  3. Burstable (t3) slightly slower - But only by ~300ms
  4. Newer generations not faster - m7i same as m6i

Raw Data

Instance Type        API    Pending     Boot    Total
-------------        ---    -------     ----    -----
t3.medium          3718ms     16620ms    2744ms   23082ms
t3.large           3885ms     16457ms    2979ms   23321ms
m6i.large          3727ms     16553ms    2672ms   22952ms
m7i.large          3789ms     16368ms    2584ms   22741ms
c7i.large          3581ms     16368ms    2727ms   22676ms
r7i.large          3464ms     16477ms    2700ms   22641ms