Now that you've written some awesome code, it's time to get serious. You need some massive firepower to get the job done, you need it now, but you also don't want to pay an arm and a leg for it.
Introducing Auto Scale Clusters!
An auto.. scale.. wha...?
Auto Scale Clusters in Ronin are your very own High Performance Computers that automatically scale by CPU's as workloads get more intensive.
They're really easy to fire up. No, really!
Lets get started
First, navigate to the Auto Scale Cluster page in Ronin and click + NEW AUTOSCALE CLUSTER
SELECT AN OPERATING SYSTEM
We offer up the most popular operating systems for HPC work. In this example it is Ubuntu 16.04.
SELECT A HPC SCHEDULER
Select one of the three most requested schedulers to work with. We have chosen SLURM in this example.
ADDRESS YOUR CLUSTER
Make this cluster available with a friendly address. This address will be how you SSH to your machine with Ronin Link. We will connect to example.ronin.cloud.
SELECT YOUR MASTER / COMPUTE NODES
Here you can choose your hardware. Depending on your requirements, you might want a small head node for interacting with, but large compute nodes for the processing power!
SELECT YOUR SHARED STORAGE
The shared storage will be available to both your head and compute nodes for you to write your results directly to.
SELECT YOUR SSH KEY
Secure your cluster with a SSH Key. Here you can create a new key, upload a key, or select a key you have used previously.
SELECT THE SCALING OPTIONS FOR YOUR COMPUTE NODES
This option allows you to fine tune the way your cluster performs under load.
- Min Machines (0 - 120) - How many machines you want running and ready when work is distributed to the cluster.
- Max Machines (1 - 120) - How many machines you want your cluster to scale to under load.
- Scale by X Machines Under Load - Once all compute nodes reach 90% of their capability, this number determines how many compute nodes to increase by.
USE AMAZON SPOT MARKET (OPTIONAL)
The Amazon Spot Market is a way for you to score cheaper hardware, by bidding for unused resources.
On Demand - The retail cost price of your hardware. It is available 24x7 and will always be there when you need it.
Spot Price Bid - The bidding price you're willing to pay for each compute node. It is subject to availability, but if your code can handle lost compute nodes, you may get the job done for up to 9 times cheaper!
REVIEW CLUSTER POTENTIAL
A quick summary of your cluster's potential!
In this example, for your very own HPC Cluster (this one is fairly tame) it will cost 6 cents per hour while you're not using it, but when you've maxed it out, you have 121 CPUs at your disposal!
REVIEW FORECASTED SPEND WITH CURRENT SETTINGS
At the bottom we try to show some estimated costs based on typical usage. This is so you know how much it will cost before committing, and prevent a nasty bill you weren't ready for.
Whew! That was a lot to take in.
Once you're happy with your choices, hit the LAUNCH MY CLUSTER button!
Return to the cluster summary, and you will see your new cluster come alive in a few minutes.
And when it does, you will see a card like this!
Well done! Look out for our next article on Cluster Tools.