Linux management software developers have taken wrong turns, and Linux admins have suffered the consequences. As a result, managing Windows and Unix is easier than managing Linux today, according Matt Mosman, CEO of San Francisco's Levanta, formerly LinuxCare.
Levanta is unveiling a Linux systems management appliance at next week's LinuxWorld Conference & Expo in San Francisco. To set the stage for the introduction, Mosman explains the differences between managing servers on Linux and managing them on other platforms, and why Linux has lagged behind Windows and Unix in server management.
Why haven't IT shops been able to just use the same management tools for Linux that they've used for Windows and Unix?
Matt Mosman: Linux deployments typically involve many more server instances than Unix or Windows systems, and the cost of managing each server goes up as more servers are deployed.
While the Lintel hardware and software cost structures are attractive, deploying 10 times as many cheap Linux boxes will cause an overwhelming administrative burden. The proliferation of Linux servers causes a problem of abundance: The more servers in use, the more differences and interdependencies. Even if a set of systems is functionally equivalent, they are never complete clones -- each system is slightly different.
Each of these differences, even if just a line in an "etc" file, must be managed and maintained over time. All of this is true whether you plan to run Linux on a server farm, a blade server, a virtual server (VMware or operating system partitions) on an SMP box, or even a mainframe. The heterogeneity of Linux creates management scenarios that are unique to the OS and different from Windows and Unix management challenges.
How has the fact that many companies now run Linux and Windows (and sometimes Unix, too) complicated server management?
Mosman: Companies now must make the decision of whether they want a single, holistic management tool that's mediocre, at best, at managing each environment -- or go with best-of-breed management tools that are superior at managing one environment, but not capable of managing others. Levanta believes that the later -- the best-of-breed approach -- is preferable by far. There isn't yet a commercially available tool that manages all OSes efficiently ... don't believe the marketing hype.
So how do you sum up today's state of processes and products for Linux management?
Mosman: Although Linux management complexity has been growing, the administrative tools and best practices have been relatively stagnant. Most Linux systems are administered through a series of scripts and freeware that are very flexible, give good "hands-on" control, but are very time consuming.
Scripting and procedural administration that grew up in the past are not up to the task of efficiently managing hundreds or even thousands of diverse servers. The situation gets even starker when an IT operation wants to repurpose servers on the fly to accommodate changing workloads.
There currently aren't any sophisticated open source change management tools. There are open source patch management and monitoring tools, but none that do total change management across the entire system.
Could you offer an example of a Linux server management method that's being used today?
Mosman: Referring to change and configuration management, an image-based approach has been used, where you use a product like Symantec Ghost, or what Dell does when they manufacture a machine. You can make an image of the whole file system and clone the drive.
That works reasonably well if you're doing mass production. But if you're trying to use it as a mechanism for change going forward, it has a lot of disadvantages. Trying to do things incrementally means you have to flush everything out, wipe the slate clean and start over.
What's another method?
Mosman: The other traditional approach is the procedural approach. Procedural approaches can be implemented by a software product, or they can be stitched together. In either case, they automate the installation and change processes that is built into either the OS or the particular application or patch being deployed.
Because they go through all the usual setup steps, in terms of sheer speed, it's only a little faster than doing it manually, although it does free up the administrator's time. More importantly, with a procedural approach you only know about things that happen to the machine when they go through the machine's procedural gatekeeper.
Another problem is that you end up with nested layers of complexity that branch over time into complex procedures. Trying to go back in time and undo something, or trying to return to a previous state, can be problematic.
These factors combine to make procedural approaches unscalable as the number and heterogeneity of managed machines increases.
Is there a common failing in both of these approaches?
Mosman: In both image-based and procedural approaches, the hardware is wedded to the OS, which is the way things are normally done. Our state-based approach doesn't require going through all those procedures. We capture the state of a machine and everything about it using our MapFS technology to keep track of the entire file system.
Do you think that a big barrier to effective Linux management has been IT staff's inexperience with and lack of training on Linux?
Mosman: Yes. Why is this the case? In large part because the management tools commonly used in Windows have a lower skill-set barrier to entry than those used in Linux. Add the diversity of Linux species to the mix and the "heterogeneity problem" makes the management challenge even more complex.
Further, the ubiquity of Windows has created an ecosystem and knowledge base for administrators to quickly get up to speed on the related skill sets. As Linux continues its penetration into enterprise, and Linux continues its popularity at educational institutions, the scales will even out a bit.
It's worth noting that the inexperience and lack of training for Linux apply doubly to SMEs and SMBs [small and midsized businesses]. Smaller companies just don't have the resources or staff to address a lot of the Linux support issues. And the tools that the smaller shops have at their disposal are also much cruder. SMEs and SMBs certainly aren't running Tivoli or CA Unicenter to manage their Linux environments. The small shops that are running Linux are doing so in a very manual, labor-intensive way.
Levanta will be introducing an appliance for Linux management. Why do you think that the appliance model will work when others haven't?
Mosman: An appliance drastically reduces the time to implementation, and makes the Linux management software more accessible to junior level systems administrators. Instead of having to manage separate entities of storage, Linux OS and the management system, you have a simple appliance with an intuitive interface.