(The following article was contributed by Micah Waldman, Vice President, Product Management and Marketing, Scali, www.scali.com. He addresses the topic of Total Cluster Lifecycle Management in the Linux Datacenter, with a focus on Optimizing Investments with Enterprise-Class Discipline.)

It’s official – Linux clusters are the new hotbed of enterprise computing. According to analyst firm IDC, in 2005 deployments of Linux-based clusters for high-performance computing surpassed those of traditional SMP machines, a sure sign that cluster computing is rapidly becoming an enterprise reality. But as Linux clusters move from high performance computing (HPC) and academic environments into enterprise datacenters, the price/performance benefits of commodity computing can be quickly eroded by the snowballing cost of cluster lifecycle management. Not only is cluster deployment still too difficult, but also the ongoing management of multiple clusters throughout their entire lifecycle is often not addressed. To extend and optimize the investments enterprises are making in Linux clusters, a new management approach is required.

Linux grows up – or does it?
Although Linux clusters are becoming mainstream, Linux systems, in general, are not managed at the maturity level of traditional Unix platforms. Managing Linux clusters is critical since they tend to multiply and quickly become heterogeneous. Management tools are just emerging for this challenge, particularly for multi-vendor, multi-cluster environments.

Tools delivered by hardware vendors are mostly insufficient for heterogeneous environments, as they typically only support their platforms. In contrast, dedicated management tools provide a centralized management solution that can replace multiple vendor-specific tools, thereby increasing productivity and service levels, and enabling automation.

Six steps toward total cluster lifecycle management
Lifecycle management is not a new concept - in enterprise datacenters, it is already applied to most mainstream IT resources. In order for Linux clusters to achieve true enterprise status, they must be managed with tools that offer a holistic approach, covering all the steps of traditional lifecycle management.

• Step 1 - Cluster Deployment: The challenge starts with quick and effective deployment. The management tool should simplify and automate the initial deployment process, from "bare metal" to a functional cluster, across multiple hardware platforms. Deploying individual servers is not enough – a fully functional cluster has to be deployed correctly including interconnect, storage, and cluster services such as NIS, NFS, and workload management. A tool that guides the system administrator (SA) through the entire deployment process ensures the implementation of best practices for cluster design, eliminates human error, and reduces time to production for new clusters or cluster expansions.

Two basic technologies can be used for server deployment. The first, image-based installation, captures a “golden” image from an existing server and deploys it on other servers. This method scales well for large clusters, provides consistent node configuration, and enables quick rollback to a previous configuration if something goes wrong. The second method, package-based installation, installs all the needed packages directly on each server. This method offers more flexibility by automatically handling multiple hardware configurations, without requiring multiple images to be created and saved. Additionally, it supports live installation of OS upgrades or security patches, without requiring a reboot.

A few cluster management tools offer intelligent provisioning by combining both methods. These tools can, for example, build the initial server configuration and then capture a “golden” image that is used to deploy other servers. Optional services such as NFS can be installed as packages on top of the base image, without requiring multiple different images. Software updates are installed as packages to avoid reboots, yet images can still be captured of new configurations for backup purposes. Thus, intelligent provisioning shortens deployment time and increase cluster uptime.