Type of Cloud
Continuing this theme on “Are you ready for the Cloud”, we are going to move forward with a new question: What type a cloud? That can be encompassed with many different connotations. It could mean it’s going to be hosted by a provider, or is it going to be an on-prem cloud? It is it going to be managed or is it un-managed? It could be a PRODUCTION cloud or is it a DR or a DEV cloud?
Today, we are going to talk about a new type, one based on Function. How is that possible? Some clouds are actually designed for a specific purposed. It’s easy to say that it is a PRODUCTION cloud or a DEV/TEST Instance. But what if you need a cloud to perform a certain way for only the applications that get assigned to it?
Take this for example: You need a cloud that can out perform your other clouds because you are going to move some high IOP dependent systems to it? What do you do? How do you build that cloud out to support what you need?
First, you need to have control over the whole cloud, not just the instances. You can create individual high-speed instances if you use a hosted automated system, but if it is going to be an on-prem cloud, then you will need to have control over it.
The high-speed cloud is broken down several ways, but we are going to focus on five: Compute, RAM, Network, Local Disk and Remote Disk. So, lets break it down:
- Compute is easy it’s your CPU Cores. At one time we would just say CPUs, but now a days we know the Cores are everything
- RAM is RAM. The faster you get, the better you are
- Networking is HUGE! If you are building your cloud in a corporate environment, you should be taking advantage of the fastest network speeds available. If you have 10GB Ethernet connections, then you should be using a few for every Compute Node.
- Local Disk is important also. Not only do you need it for your OS, but also for direct storage capability. Local storage is handled differently, so you should be aware of that. If you 15K disks spinning in the Compute nodes; that will give you a great start for simple storage, as long as you watch the sizes of the volumes you create.
- Remote disk can be a corporation’s saving grace. Local storage can fill up fast, and you are limited to the number of IOPS your disks can spit out. Going with a remote flash/SSD system can get you upwards of 500K IOPS per second. WAY faster than a local disk.
So you start assembling your cloud. You take several Compute nodes, and you load them up with CPU cores, RAM and high-speed local disks. Then you have your Compute nodes attach to the remote disks (or sub systems) to create and provide the big, fast chunks of storage.
Now, when you send your high-speed applications to it, you separate out the application servers and database servers within the cloud so the Compute nodes can focus on running one task, instead of hundreds.
Can you add other server instances to the Compute nodes? Yes, but make sure you do some system tests first to make sure you are not using to much horsepower for one application, and then starving another.
By Richard Thayer