Cloud, Private Cloud, & Bare Metal Resources

Last Updated October 2014

Pagoda Box gives you the ability to easily grow your infrastructure as the demands of your app grow. As traffic and/or demand increases, you have the ability to scale any of your services to three different levels of resources, depending on what you need: Cloud, Private Cloud, and Bare Metal. This doc will walk you through the specifics of the three. After reading it, you should be familiar with:

  • Key characteristics of Cloud, Private Cloud, & Bare Metal resources
  • Which type of resources will best fit your needs
  • How to grow your app to a massive scale with a few button clicks

Cloud Resources

Cloud resources are the foundational unit of Pagoda Box. They are small, easily scalable and deceptively powerful. Everything on Pagoda Box has been optimized for performance, so don't be surprised when your apps perform better on Pagoda Box running on less resources than on other hosts. With cloud resources, you can expect the following:

Dedicated RAM

RAM for Cloud resources, at any scale, is dedicated. Your service is guaranteed access to the amount of RAM it's scaled to. For services with multiple instances, each instance has its own dedicated RAM. Instances don't share RAM. For example, if you have a web service with 3 micro-cloud instances, each instance gets 128MB of RAM. The 128MB are not shared between the 3.

Shared CPU, Disk IO, & Network

Cloud services share CPU, Disk IO, and Network with other cloud services on the same physical machine. It's important to note that even though these resources are shared, there are load-balancing systems in place to ensure every service gets the resources it needs. But there may be some additional queuing that happens on cloud resources that you wouldn't see on Private Cloud and Bare Metal Resources.

CPU Availability is Based on Priority

CPU usage for cloud services is based on priority and is burstable up to 4 Core 2 Duo cores. The best way to describe how CPU priority works is to compare it to a lottery. Each time a process requires compute time (CPU) it puts a ticket into the lottery. When a CPU core is available, it pulls a ticket out and the process to which it belongs goes through next. A "5x" priority is like putting in 5 tickets in the lottery, making that request 5 times more likely to get CPU time than a 1x priority request. Keep in mind this happens extremely fast (on the scale of microseconds) and for the most part, won't result in noticeable latency unless your service is CPU intensive.

Private Disk Space

Every Cloud service has its own private disk space.

Private Cloud Resources

Each Private Cloud service and/or instance is basically a Virtual Private Cloud Server. They're perfect for services that need more power than cloud resources can provide. With Private Cloud Resources, you can expect the following:

Dedicated CPU, RAM, & Network

Private cloud resources each have dedicated use of the specified number of CPU cores, amount of RAM, and Network. They aren't shared with any other processes running on the same physical machine.

Shared Disk IO

Because Disk IO is a server-level metric, it is shared between all processes running on the same physical server, including Private Cloud services. If your service depends on high IO, Bare Metal resources may be a better solution.

Private Disk Space

Every private cloud service has private disk space.

Bare Metal Resources

With Bare metal, you get your own dedicated machine. No sharing of resources whatsoever. Bare Metal resources are the power-houses of the Pagoda Box arsenal, providing the highest level of processing power. With Bare Metal, you can expect the following:

Private, Dedicated Everything

Nothing is shared with Bare metal resources. They are physical servers, so whatever specifications you specify are totally dedicated to your service. CPU, RAM, Disk IO, Network, and disk storageā€¦it's all yours.

24 Hour Provisioning Time

Bare Metal resources typically take a few hours to provision but can take up to 24 hours. Service creation and scaling transactions will take at least this long.

If you have any questions, suggestions, or corrections, let us know.