Hyper.sh

» see hyper.sh as it lived, on the wayback machine

~ the goodbye ~

If you ran Docker containers in 2017 you had two shapes of bill in front of you. You could rent a virtual machine by the hour and run your containers on it, paying for the machine whether anything was running or not, or you could use one of the early container services that were mostly virtual machines with the renting hidden from you. Hyper.sh was the one that actually did the thing the pitch described. It booted your container straight onto a hypervisor with no operating system in between. You paid per second of runtime, and nothing at all when nothing was running.

This was my territory, unlike most of what is in this archive. I am an engineer and 2017 was the year Docker stopped being a thing people argued about and became a thing people just used, and per-second billing on a container was the sort of number you read twice. The technology underneath was open source: a daemon called hyperd and a runtime called runV, built so that a container got its own kernel and its own isolation boundary rather than sharing the host's.

~ the chronicle ~

~ what per-second billing meant ~

The pricing was the product. Cloud providers of that era sold you time on a machine, and the gap between what you paid for and what you used was where their margin lived. Hyper.sh priced the container itself, so a job that ran for nine seconds cost nine seconds. They opened a European region in Frankfurt in September 2017, people wired it into Jenkins and into build pipelines where work arrives in bursts, and in 2017 an independent cost comparison put it at the best value of the hosts tested.

~ the pivot ~

In December 2017 runV was merged with Intel's Clear Containers project. The combined thing became Kata Containers, hosted by the OpenStack Foundation under Apache 2, and it is still alive today in a way Hyper.sh is not. ARM announced joint engineering work with the Hyper team on the runV project. The open source side of the house was winning.

The hosting side was holding on, but only just. A small company selling infrastructure against Amazon and Google has to be meaningfully cheaper forever, and forever is expensive. By late 2018 the team had decided that its future was the Kata project and an enterprise Kata stack sold for deployment on the major public clouds. They would be shipping onto the infrastructure of the competitors they had spent three years undercutting.

~ the day it died ~

The customer email went out on 21 December 2018, four days before Christmas. It said that creating a new account was already impossible, that the service would stop on 15 January 2019, and that any remaining user data and accounts would be deleted the same day. It cited section 11 of the terms of service. Nobody quotes a clause number at customers until they have decided the conversation is finished.

Their own blog then posted "Farewell from Hyper Crew", and that post says the platform was officially shut down as of 16 January 2019. One day past the announced date, which is a small thing, except that the announced date is what every secondary account of this repeats and the farewell is the only source written after the fact. In March 2019 hyper.sh stopped resolving to anything at all. The domain is dead, the daemon is archived on GitHub, and the kernel-isolation idea Hyper spent four years on is now running inside other companies' products under a name they gave away.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »