360 words, about 1 minute.
The site is built and maintained by me using Hugo, a static site generator written in Go. It’s running on a FreeBSD nginx web server. By the time you read this nginx will probably be inside a jail. The theme is a port of Ghost/Casper. My criteria was low maintenance, good security, performance, and simplicity. The content is all written in Markdown in a text editor, Hugo’d, and uploaded to the server using rsync/ssh. The source tree lives on cloud storage and takes up 18M of space. The FreeBSD server has 256MB RAM. Comments are managed using Disqus. Let’s hope it keeps the spam down and they do a better job of security than they have done in the past. The entire stack is updated daily from ports, freebsd-update, and Certbot/Let’s Encrypt. We’ll see if it is simple enough to take this rough treatment.
So far it has been up 182 days, I haven’t physically lifted a finger to do maintenance for 719 days. These figures are updated regularly and automatically.
Posts are made from a desktop or laptop, but with a little extra effort I could do the same from a phone or tablet. Maximum simplicity, minimum attack surface, minimum maintenance. Why don’t I simply use built-in integration of CoSchedule together with Evernote? In a word, WordPress. While WordPress is a popular and featureful content management system, and Evernote->CoSchedule->Wordpress works, it requires more ongoing effort to keep secure than I want to spend. I don’t want to be informed via twitter that I’m hosting malware, phishing, or any other undesirable content (unless I put it there). I want a set and forget arrangement. Things which are currently missing from the stack include CDN, DDoS mitigation, and anycast DNS, none of these are critical for now. The biggest drag on performance is the Disqus feature and its various beacons and advertising-related trackers.
Today the site is a self-hosted Virtual Machine but could just as easily be deployed as an Amazon Lightsail or some other inexpensive VPS.
This tiny VM has 1000x to 10,000x more performance than the 1960s monster pictured above. Cost per unit of performance is 1000x less. I’ll return to these numbers in a post about Artificial Intelligence, and why you shouldn’t listen to academic professors predictions about the future.