Performance Dynamics Company

02/22/2026

Does the LLM efficient-compute frontier represent new physics? — Instantons in the machine https://speakerdeck.com/drqz/does-the-efficient-compute-frontier-represent-new-physics

11/12/2025

In the vein of: The cloud is the new mainframe and Charge-back is back, baby!
https://speakerdeck.com/drqz/how-to-scale-in-the-cloud-chargeback-is-back-baby


just launched 'Instance Explorer': a simple way to search & compare running instances (including forgotten ones) on AWS, Azure & GCP https://www.linkedin.com/feed/update/urn:li:activity:7394024965781762048/

💸 “Am I overpaying for this instance?” Definitely in the Top 10 most overheard comments in cloud engineering teams, right between “Who spun this up?” and “Do we still need this?”. Now there’s a free tool to actually answer that - built by Datadog. We just launched Instance Explorer:...

10/23/2025

Ordinary AI output requires extraordinary human verification. (Apologies to C. Sagan)

10/22/2025

10/20/2025

DR still in progress for AMZN.com itself ... 🧐

10/13/2025

Productivity: If you want to be more productive, go to sleep.
https://www.perfdynamics.com/Manifesto/gcaprules.html .25

10/05/2025

Some performance consequences often overlooked in redundancy plans
https://perfdynamics.blogspot.com/2014/03/performance-of-n1-redundancy.html

09/18/2025

Out of The Queue Comes ... Amdahl’s Law!
https://www.perfdynamics.com/Papers/amdsynq.pdf

08/22/2025

A clue for remembering all THREE of Little's laws, due to Guerrilla grad Jeffery Pickett

See https://perfdynamics.blogspot.com/2016/10/a-clue-for-remembering-littles-law.html to find out the meaning of Q, L, U, and R, W, S.

08/08/2025

RIP James Lovell, commander of Apollo 13 (played by Tom Hanks in the movie of the same name), has died at age 97.

08/07/2025

Displaying metrics, errors, etc. ... is not the same thing as displaying metric errors.
Simple example. Consider Unix, Linux, macOS 'top' output showing CPU utilization.
1. How it is (old school)
CPU usage: 5.79 %usr, 4.51 %sys, 89.68 %idl

These data are sampling averages.
I'm ignoring the number of significant digits here
https://lnkd.in/gP96NDim

2. How it could be (nay, SHOULD be)
CPU usage: 5.79 ± 0.29 %usr, 4.51 ± 0.23 %sys, 89.68 ± 4.48 %idl

All measurements (data) are wrong ... by definition.
The only question is, can you tolerate the wrongness?
If you're not informed how wrong data is, how can you decide?