Hyper.sh: Container-native Cloud

https://docs.google.com/presentation/d/1nrfEcCITAofSDPFTIkiASLa4UvmCOUmgPE6ayFZBR0c/edit?usp=sharingl

by scprodigy • 9 years ago

99 29 9 years ago

docs.google.com

alauda• 9 years ago

Video is here: https://www.youtube.com/watch?v=JCldsRNuGkc&feature=youtu.be

voiper1• 9 years ago

https://www.hyper.sh/pricing.html

  Cheapest single container with a publicly accessible IP (FIP) assuming a 1GB Docker image size.
  S1 container: 1.03/month
  FIP: $1/month
  RootFS Container Volume (10GB@$0.1/GB/month): $1/month
  Container Image (1GB@$0.1/GB/month): $0.1
  Total monthly price: $3.13/month

initdaemon2• 9 years ago

They have per-second billing. Use it as serverless.

tyingq2• 9 years ago

5 second boot time though. Better than a normal VM, but maybe not great for user facing serverless.

e12e• 9 years ago

I'm intrigued by network usage being "unlimited". Does this mean unlimited streaming of 10GB of video for a little over 3 USD/month?

mjmein• 9 years ago

I'm interested - what type of app can you run on 64mb of ram? What languages are suitable for building these types of microservices?

Gaelan• 9 years ago

I’m pretty sure most languages, even interpreted, can run at least small programs in 64MB. If you want to minimize memory usage at all costs, I think you’d want a non-GC compiled language (C(++), Swift and Rust come to mind).

lobster_johnson• 9 years ago

You can forget about Node.js, Python or Ruby for any meaningful app, like a web API. Erlang/Elixir is also iffy, I'm pretty sure. Java, last I check, had a default heap size of 64MB, and does badly on less.

I've written small Go programs that can consume around 15-20MB on low traffic. But they also have huge spikes in memory usage when you have concurrent requests hitting them.

curun1r• 9 years ago

I'd just like to say how dispiriting I find this question considering I how envious of my friends $4k+ computer I was because it had 64mb of RAM and could play C&C Red Alert with absolutely no slowdown. Let that sink in...in the mid 90s, we comfortably fit a Windows GUI and a somewhat sophisticated game into 64mb of RAM and 20 years of "progress" later, we're asking what kind of special accommodations are necessary to handle HTTP requests with the same resources.

Note: It's an entirely reasonable question to ask, which makes it all the sadder.

provost• 9 years ago

My favorite comparison is the Apollo guidance computer's memory (though completely different comparison from your GUI and game description). But the AGC is absolutely fascinating.

16-bit wordlength, 2048 words RAM (magnetic core memory), 36,864 words ROM (core rope memory)

Source: https://en.wikipedia.org/wiki/Apollo_Guidance_Computer

meddlepal2• 9 years ago

Thousands of HTTP requests per second. Not just "a handful".

ceronman• 9 years ago

The original PlayStation had only 2MB of RAM and 1MB of video memory. Still you could play games such as Metal Gear Solid on it :D

solatic• 9 years ago

It's rather common for managers to want simple, human-readable reports for information which is already exposed through a REST API.

Setting up a container with a cronjob that, once per day, executes a simple shell script which curls a REST endpoint, parses the relevant information, and sends an email out to a pre-defined mailing list or puts it to a slack channel... Yeah, that's pretty simple to pull off within that memory footprint.

drej• 9 years ago

nginx

deforciant• 9 years ago

Go is a good example - I have several services that consume ~20MB of RAM so no need for beefy boxes. Although that's why I also chose to use Kubernetes - extremely easy to pack a bunch of services.

sawantuday• 9 years ago

I have a PHP based crawler scripts that range from 30 to 55 MB

NathanKP• 9 years ago

A Node.js REST API can handle a significant volume of incoming requests with only 50mb-75mb of RAM. For reference here is a load test of handling around 75k-80k reqs/min using node.js containers deployed using Amazon ECS. Each node.js process is basically just doing CRUD operations, accepting incoming JSON, storing it in a database, fetching values back out of a database, and sending JSON back to clients. There is a bit of extra overhead from microservice HTTP fanout to other backend microservices, but the whole stack is very lightweight and 99% of requests are served in under 30ms.

As you can see this workload is only using about 6% of the available memory. For event loop driven systems like Node.js or Nginx, or Go the workload is almost always CPU bound not memory bound:

http://i.imgur.com/VCN5e21.png

jamesrom• 9 years ago

HyperApp: https://itunes.apple.com/app/apple-store/id1179750280?mt=8

Hyper.app: https://hyper.is/

hyperapp: https://github.com/hyperapp/hyperapp

Hyper.sh: https://docs.google.com/presentation/d/1nrfEcCITAofSDPFTIkiA...

Adding this to the list https://news.ycombinator.com/item?id=14634284

lucideer• 9 years ago

There are countless examples of overlaps in software naming. I'm sure this isn't even the word with the most duplicates.

Also, the .sh does make this particular example at least slightly distinctive.

OJFord• 9 years ago

Hyper.is is a Mac terminal. I used it for a time, and would always go to hyper.sh looking for docs or whatever.