Editor’s Note:
Dell Solution Centers in conjunction with Intel have established a Cloud and Big Data program to deliver briefings, workshops and Proofs of Concept focused on hyper-scale programs such as OpenStack and Hadoop. The program’s Modular Data Center contains over 400 servers to support hyper-scale capability to allow customers to test drive their solutions.
In this blog series, Cloud Solution Architect Kris Applegate discusses some of the technologies he is exploring as part of this program – and shares some really useful tips! You and your customer can learn more about these solutions at one of our global Solution Centers; all have access to the Modular Data Center capability and we can engage remotely with customers who cannot travel to us.
*******************************************
With the tremendous success of virtualization, the need to create larger and larger compute clusters is become necessary. To support this scale, building large elastic pool of servers can provide a great way to scale private clouds up and down as demand dictates. VMWare has provided a toolkit with VMWare vSphere Enterprise Plus called Auto Deploy that assists with just that.
Auto Deploy allows one of two methods to deploy your ESXi Hypervisor:
Stateless– This is where the OS is streamed over PXE / TFTP to each compute node. In this case, you could have little to no storage hardware local to the hypervisor node. You can create a fleet of compute-only nodes that have SAN adapters, Network Adapters, Memory, and CPUs that can provide all the necessary compute services to satisfy the Virtual Machines.
Stateful– This is where you use the same PXE / TFTP mechanism to initially install the ESXi OS to local media (either SD Card or local Disks). Once the installation is done, the Auto Deploy mechanism disengages and local boot of the OS occurs moving forward.
Recently, we had the need to spin up a very large VMware Big Data Extensions cluster of 100 physical nodes. These nodes all had a large amount of physical disks (24 local disks per node) that we wanted to be able to dedicate to Hadoop and not have to use them for local OS data. This became the largest known physical cluster of this type to date.
Image may be NSFW.
Clik here to view.
The high-level process for setting up stateless Auto Deploy is as follows:
- Install and Setup VMWare vSphere 5.5 Virtualcenter Server (VM or Physical).
- Install Auto Deploy Server (VM or Physical).
- Setup PXE and build a “First Boot” profile for the first stateless node.
- Boot the first node and make sure that it joins the Virtualcenter.
- Customize the server’s network, datastores, and other configuration parameters that will serve as a template for all other nodes.
- Create a Host Profile based off this server.
- Create a new Auto Deploy profile based off this new Host Profile.
- Boot all your servers and watch as they mimic the configuration of the previous host.
Rolling out a massive amount of servers in a repeatable and scalable manner is critical for enabling private clouds.
Image may be NSFW.Clik here to view.
