Slashdot Mirror
Google Offers Cheap Cloud Computing For Low-Priority Tasks
jfruh writes: Much of the history of computing products and services involves getting people desperate for better performance and faster results to pay a premium to get what they want. But Google has a new beta service that's going in the other direction — offering cheap cloud computing services for customers who don't mind waiting. Jobs like data analytics, genomics, and simulation and modeling can require lots of computational power, but they can run periodically, can be interrupted, and can even keep going if one or more nodes they're using goes offline.
cheap cloud computing for customers. called it!
If only I was not using the low-priority cloud compting..
It sounds like AWS's Spot Instances? Except for the fixed pricing.
Yup, it's their version. Forbes compares them. The fixed price is nice on the Google side, but there's no 2-minute warning before termination on Google like you get on AWS, and AWS launched a new Spot Fleet product the same day Google announced.
Either way, you need to be doing the kind of work where you can lose VMs on short notice and keep going, but it's a very nice discount if you can.
Socialism: a lie told by totalitarians and believed by fools.
They do give you 30 seconds of warning: https://cloud.google.com/compu...
Compute Engine performs the following steps to preempt an instance:
Compute Engine sends a preemption notice to the instance in the form of an ACPI G2 Soft Off signal.
If the instance does not stop after 30 seconds, Compute Engine sends an ACPI G3 Mechanical Off signal to the operating system.
Compute Engine transitions the instance to a TERMINATED state.
So if you're able to persist your state in less than 30 seconds, just watch for SIGTERM and you should be golden. Otherwise, checkpoint frequently.
The cloud has never been anything but old technology with new buzzwords.
I do not fail; I succeed at finding out what does not work.
i know you guys used to have operating systems and a process abstraction. but now i nest operating systems and processes
layers deep, and went back and added lots of hacks to try to mitigate the huge overheads involved with doing so. but
since that wasn't enough indirection, i added a container layer on top of all that.
my container runs on a virtual network which is attached to another virtual network
none of this runs without a bevy of distributed services that i don't even know about, much less understand
now i can't say anything about anything, and when it fails i simply shrug
bow before my awesome power you neckbeard
This is a really good idea from a resource utilization standpoint. Reminds me a lot of the power grid - you have to have enough hardware to handle the peak load, but idle capacity just costs you money.