Distributed Image Management for Linux Clusters
A scalable image management tool that allows blades to run a Linux distribution over the network without a local disk.
Date Posted: April 24, 2007
|
|
 |
|
|
|
Distributed Image Management for Linux Clusters provides all tools necessary for setting up and maintaining large and very large clusters, especially clusters based on (but not limited to) IBM® BladeCenter®. The main subsystems managed by Distributed Image Management for Linux Clusters are the following:
- IP addresses
- DHCP
- NFS
- file system images
- network boot images (BOOTP/PXE)
- node remote control
| | |
|
No. The primary focus of Distributed Image Management for Linux Clusters is on managing the Linux distribution image for all the nodes of the cluster. It turns out that this is often one of the most difficult cluster management processes. Distributed Image Management for Linux Clusters can be complemented by other cluster management suites. | | |
|
Not exactly. The nodes in the cluster do not require disks. However, your images must be stored on the disk of the image server. | | |
|
No. A RAM disk is only used for booting. Early in the boot process, Linux will change root (chroot) to a comprehensive, network-based root file system. Therefore, Distributed Image Management for Linux Clusters is unlike embedded Linux installations with minimal function. On the contrary, users of Distributed Image Management for Linux Clusters often have large RPM installations in order to support many functions of a wide user base. | | |
|
Stateless operation is a desirable cluster policy for many reasons. However, Distributed Image Management for Linux Clusters will preserve state as necessary. Therefore, a stateless policy is not required. That is, changes to the read/write portion of the root file system for any particular node will be preserved between boots of that node when the node is booted to the same image. | | |
|
Currently, all the nodes connected to this image server will stop and must be rebooted. | | |
|
- Distributed Image Management for Linux Clusters provides a very flexible mechanism for defining the IP address layout of thousands of nodes in a few lines of XML.
- Distributed Image Management for Linux Clusters also has some useful command line tools for administering Blades in IBM BladeCenters or rack-mounted servers via the Remote Supervisor Adapter (RSA) and via SNMP.
- In Distributed Image Management for Linux Clusters, one can set up multiple images for every node in parallel. For example, one can set up images with RHEL and others with SLES9, even in advance without having the nodes currently available. Switching between these images takes only two Distributed Image Management for Linux Clusters commands: one to change the MAC addresses in DHCP and one to reboot the nodes.
| | |
|
Currently, there is no limit. Distributed Image Management for Linux Clusters can be scaled very well because of its modular design. Today the largest cluster managed by Distributed Image Management for Linux Clusters has 2500+ nodes; however, there shouldn't be any reason to go much beyond this number. | | |
|
Depending on the number size of the files, it can take a few seconds or up to less than one minute. | | |
|
It might be more difficult to build the initial set-up compared to other cluster management systems. Distributed Image Management for Linux Clusters has been designed for large and very large clusters and cannot fulfill all possible requirements. For example, every new Linux distribution requires some adjustments of the Distributed Image Management for Linux Clusters code. But, as a result, you will get a very efficient and scalable system. Skills in a scripting language such as Perl and Bash are definitely helpful. | | |
|
No. | | |
|
We are planning a fault-tolerant image server so that if the image server goes down, the nodes will remain active. | | |
|
You need RedHat, SLES, or Fedora Core installed on one node in your cluster. This is called the master image. You also need an image server with enough disk space to contain at least two copies of the image. We recommend at least 20 GB + 0.3 GB per node of free disk space on its own partition. For very large clusters, you will need to plan the number of nodes you want to support and the number of versions of a distribution you want to maintain. You also should have some plan on how you want to organize IP addresses in your cluster. See the examples on how to define IP addresses in the conifg directory (/opt/dim/config).
Linux is a registered trademark of Linus Torvalds in the United States, other countries, or both.
IBM and BladeCenter are trademarks of IBM Corporation in the United States, other countries, or both.
Other company, product, or service names may be trademarks or service marks of others.
| |
|
|
|
|
| |
