About notP
Introduction
notP is a loose coterie of applications that, when combined, provide a simple means for the administrator to quickly force nodes to boot into various OS images. While it can be used as a stand-alone application, notP was designed to be used by Moab/Maui and Torque in a HPC environment, quickly provisioning nodes per user demand.
Status
notP is alpha stage software. While we use it in production ourselves and it works great, there are no guarantees that your experience will be as plesant as ours. Ideally, you will find at least the underlying idea interesting and worth enough of your time to send us comments, suggestions, or at the very least, any problems you experienced so that we may do our best to improve.
Currently, we are focusing on documentation and getting this project out in the open (Thanks Sourceforge!). While this is going on, we're also closely watching what is happening with notP on our Xserve cluster. Of course, there's quite a number of features we like to implement, and we'll have not problem keeping busy. Still, it's our plan to wait and see if anyone has any thing they'd specifically like done. So let us know!
Why?
I can think of two reasons off the top of my head:
- High Performance Computing Clusters: On site we have users who prefer OS X over Linux and vice versa. Historically, we separated the clusters using vlans which required mirroring of all the services (scheduler, resource manager, LDAP, BIND, etc) across the systems. Then we got smarter and separated the machines using the Torque pbs_mom arch string. The problem essentially was that we had a hard number of OS X and Linux nodes, which lead to an under-utilization of resources. With notP, the number of OS X or Linux nodes changes automatically depending upon user request.
- Education Computing Clusters: Simply put, you can have a class using OS X followed by a class using Linux on the same machines. In the future, we plan on playing with Windows as well.
How?
- Blancmange
Blancmange, as the name hints, describes a number of parts of notP. Initially, one meets Blancmange when building an initial initramfs used to boot nodes via PXEboot or Yaboot. It is a collection of scripts that download, build and install into a small image the required tools. The scripts also take care of building the rest of the minimal filesystem.
The minimal Linux environment that is built inside the initramfs is also called Blanchmange. This initial image has only one purpose, to install itself and any other required tools (like cid, extra rc scripts...) onto the netbooting node.
Finally, for the duration of your use of notP, Blancmange refers to the notP bootloader. In other words, the small system that takes care of provisioning nodes when necessary.
- Cid
Cid is the guide of the provisioning system, operating as follows:
- Receive a request for node1 to be provisioned to cluster_os, version 1, subversion 2
- Check the status of node1. If running cluster_os-1-2 the process halts.
- Otherwise, node1 reboots to Blancmange and reports the images currently available.
- If cluster_os-1-2 is available, node1 will boot to it and the process halts.
- Otherwise, node1 will perform the steps to install cluster_os-1-0 and update to 1-1 and 1-2 as needed
- In the end, depending on the amount of work necessary to provision node1, we guarantee node1 is running cluster_os-1-2
Dependicies
Without the following packages, notP would never have existed.