Not legally an Engineer Yet, Call me an EIT
Just a page where I dabble in partially formed thoughts. Getting them onto paper, or rather Web, so that I can more fully form them.
Why a website would have operating hours is more or less beyond me. But I’ve come across a handful of them. Websites that when access outside of “Normal Business Hours” will display a simple message asking you to come back during specific hours to access the site. I think this sticks in my mind because of how fast web hosting costs can increase. And how much money can be saved by simply cutting down the hours a site is operational during. Of course you always need to have something small running to serve the bare bones come back later message, but that can be handled by an extremely small server, and if you’re doing this type of come back later you’re likely already getting the vast majority of your traffic at specific times of day.
I did try for awhile to do something by having a site which was served from a scalable instance that dropped to zero. The issue was that the site typically took far too long to boot up and people grew frustrated with the loading times being drawn out whenever they accessed a site for the first time in awhile. The issue of startup times might be addressed to some extend; however, I have serious doubts it could be gotten down below the 100-200ms that I find online is considered to feel “Instantaneous”. Although firecracker vms are supposedly able to start up in under 125ms which is potentially sufficient to acomplish the instantaneous feel that’s desired, if with minimal room. One thing I’ve considered is could a load balancer potentially allow a mix of both operating hours and faster startups. People don’t need everything to arrive all at once, they just need it all to be there pretty fast and to start arriving within that timeframe. Potentialy caching the basics of the page and serving those before the actual system has fully started up.
Unfortunately I likely won’t get to try out any of these alternate approaches as the system has settled onto a pretty good solution that fits within the needed cost point. The other issues here being that most of the potential methods for getting things setup to handle these ultra fast instant deployments are actually more expensive than simply running traditional virtual machines.
One thing that I’ve noticed as of late is the number of specialized chips that end up being more expensive than simply taking a microcontroller and programming it to fulfill the needed functionality. For instance a design I’m working on currently I need UART to USB conversion; however, instead of placing the UART to USB FTDI chip down and working with that established setup, my current plan is to take an RP2040 chip and use that instead. Essentially allowing me to also gain access to a second UART channel, as well as a plethora of GPIO. In this particular case I’m working with a low powered Microprocessor as well which uses its one USB port for OTG mode, and while it can be configured to overlap the serial port, it’s simply convenient to just add a secondary port. Worst case scenario I gain a port whereby I can interact with the RP2040 over the UART and simplify the process of controlling IO.