87 lines
3.7 KiB
ReStructuredText
87 lines
3.7 KiB
ReStructuredText
======================
|
|
Stateless Provisioning
|
|
======================
|
|
|
|
Why is Provisioning Important
|
|
=============================
|
|
|
|
Clusters are pools of servers bundled together to do a particular job
|
|
or set of jobs. While there are a number of different use cases for
|
|
clustering today, Warewulf was originally designed out of necessity.
|
|
|
|
Back in 2000, when Linux clustering was growing up for HPC, the issues
|
|
of scale became apparent. Of course in HPC, there are many scalability
|
|
factors which needed to be overcome as we continued to scale up
|
|
clusters. Pretty early on was the "administrative scaling" which is
|
|
the factor that full time systems administrators could only maintain
|
|
so many servers. While homogenous configurations were able to help
|
|
that, we still had the problem that every installed server became a
|
|
point of administration, version creep, and debugging. The larger the
|
|
cluster, the harder this problem was to solve.
|
|
|
|
Warewulf was created to help with exactly this.
|
|
|
|
Provisioning Overview
|
|
=====================
|
|
|
|
Provisioning in this definition is the process of putting an operating
|
|
system onto a system. There are many ways to provision operating
|
|
system images, from copying hard drives, to scripted installs, to
|
|
automated installs. There are many valuable tools to facilitate this
|
|
and they all helped to solve this problem.
|
|
|
|
In a cluster environment, this means one could group all of the nodes
|
|
together, to be installed in bulk. Previous to cluster provisioning
|
|
system administrators would go around to each cluster node, and
|
|
install it from scratch, with an ISO or USB thumb drive. This
|
|
obviously is not scalable. But being able to automatically install
|
|
hundreds or thousands of computers in parallel and automate the
|
|
management of these systems completely changed the paradigm.
|
|
|
|
There were several cluster provision and management toolkits already
|
|
available when Warewulf was created and while these tools absolutely
|
|
helped, there was even more optimization to be had. Stateless
|
|
computing.
|
|
|
|
Stateless Provisioning
|
|
======================
|
|
|
|
The next step past automated installs is to just skip the installation
|
|
completely; boot directly into the runtime operating system without
|
|
ever doing an installation.
|
|
|
|
This is Warewulf.
|
|
|
|
Stateless provisioning is realizing you never have to install another
|
|
compute node. Think of it like booting a LiveOS or LiveISO on nodes
|
|
over the network. This means that no node individually is a point of
|
|
system administration, but rather the entire cluster, inclusively is
|
|
administrated as a single unit.
|
|
|
|
If all cluster nodes are booting the same OS image (or set of OS
|
|
images), then any individual nodes that have problems is
|
|
hardware. Debugging software and doing system administration in single
|
|
points within a cluster is not needed. There is no version drift,
|
|
because it is not possible for nodes to fall out of sync. Every reboot
|
|
makes it exactly the same as its neighbors.
|
|
|
|
Warewulf provisions the operating system by default to system
|
|
memory. There is no need for hard drives with Warewulf.
|
|
|
|
Previous versions of Warewulf had the ability to write the operating
|
|
system to hard disk as well as do hybrid provisioning (the core
|
|
operating system in memory, other pieces overlaid over NFS) but these
|
|
have been obsoleted in Warewulf v4 as there are easier ways to
|
|
accomplish the same thing (e.g. use of swap space).
|
|
|
|
.. note::
|
|
|
|
If you wish to provision to the hard drive, we might add that
|
|
feature back based on user requests but in the mean time, you may
|
|
wish to look at an automated or scripted installation platform
|
|
instead of a cluster provisioning system.
|
|
|
|
In our experience, the Warewulf provisioning model is by far the most
|
|
advantageous, simplest, and most flexible and scalable cluster
|
|
provisioning platform available.
|