However, the code is open source and you can grab it here:
http://github.com/gerner/canonicalizable

I like the way you show how Dstat can be useful by a nice screenshot and a story. In fact something I always wanted to do was create some scenarios (and related Dstat commands) that show the reasoning and functionality of Dstat. Your example is a good one.

However Dstat is an evolving project and in the meantime we have plugins that show disk utilization rates (–disk-util) and other plugins showing more information that iostat provides as well. This would be perfect to explain you what disk is causing iowait situations.

The true strength of Dstat compared to similar tools is that it is written in python and writing your own plugins is quite easy. Which gives you a wealth of possibilities to correlate system events and counters. I am limited by my own experiences but I hope other people contribute some plugins that extend Dstat in ways that I never envisioned.

If you come across other scenarios, blog about them and send me a note Thanks again !

I’ve struggled a lot with (3). Finding (and being) a manager who can be both supportive and critical is really tough. Learning to respond to both praise and criticism as an employee or contractor is equally tough.

As I said once on Twitter – “A good manager pats you on the head and kicks you in the ass. A bad manager kick you in the head and pats you on the ass.”

nate

From the engineering side, this is probably driven by cost. What’s the cost of a page-view? Hopefully you’ve got a scalable architecture so you can say something like:

It’ll cost us $300/month to push 20 pages/second, plus ops concerns. If we want to push 40 pages/second, that’ll cost us $600/month because we’ll need to add another box.

I don’t think I’ve got a rule of thumb in general. Performance varies from application to application so much that it’s hard to say. OSE is pretty data intensive: one report issues five or six API requests for a few thousand rows of data, hits the database a handful of times, and queues up other work for background processing. On the other hand, there are plenty of benchmarks that suggest a naive ruby app that does basically nothing can do ~500 page-views/second on low-end commodity hardware.

I guess there you go: per box you should get at least 10 or 20 page-views/second at the low-end. But it’s unrealistic to expect 500 or more page-views/second from any real app on the kind of hardware you’re likely to see in production.