1. 27
  1. 6

    A big part of the cost of hosted servers is that the hardware is maintained by someone else and is located in a proper datacenter with (hopefully) fast and redundant internet links. Because they’re geared towards production services, not “lab”/staging/test/dev environments.

    However, the hourly billing that “cloud” providers offer can be useful for these environments – If you have short, rare tasks (builds, experiments) that need expensive hardware.

    Let’s say I want to compile WebKit for aarch64 every two weeks, and I want the build to be done quickly, so I need a dual ThunderX box. I could either spend a couple thousand $$$ upfront, or pay $0.1 per hour for a spot instance on Packet, which comes out to $0.2/month if the build finishes in under an hour.

    1. 5

      @arrdem, you’ve got an issue with your power calculations and/or units. Watts are units for measuring instantaneous power and Watt*hours are power over a given time. So “8kw per day” doesn’t make much sense. You might just have kW and kWh switched, I didn’t look close at the numbers. Might be why things aren’t making sense.

      1. 2

        @azdle, yeah that checks out. The numbers clearly didn’t work when I tried to do the dimensional analysis this morning, I think the mistake comes down to me recalling kwh = w/h whereas kwh = w * h which is why I went and ballparked off of PG&E’s power analytics.

      2. 2

        You might want to resize those images - you are using full size 4048x3036 images as thumbnails (which aren’t clickable anyway).

        1. 1

          @arrdem, can I ask if you use a static site generator / template for your site? If so, which one? I really like the design…

          1. 1

            I do! This site (and others) are built on Jekyll. The theme I use is a copy of the Hyde theme I’ve hacked up over the last several years. The color scheme and logo were made with https://logojoy.com

          2. 1

            It clearly heavily depends on what you need to do but UpCloud $5/mo instances work well with lots of smallish projects/websites.

            I really doubt many people would need such a big configuration for a single node and even so you might need it just temporarily (so rent it for X hours/days and then turn it off) or you might be better off getting many smaller and cheap instances to get some sweet load balancing with them.

            I’d get a on-site setup only for very specific projects and you’ll know when that’s the case, otherwise economy of scale wins (and less hassle overall).

            1. 1

              Another part of the hidden cost is replacing failing hardware as well.