5.2 Choosing the right Linux project

People tell us two things about Linux on the mainframe projects:

• It is easy to get started.

• Using a staged approach works best.

5.2.1 Start small

Our advice is to start small. The bigger the scope of the initial project, the more the entire company is involved, and the harder it will be to succeed. What would constitute a typical small project? In our experience, successful companies started with things like consolidating an infrastructure server on the mainframe or simple (static) Web serving. Choosing a low-attention project allows the team some success without the pressure of company politics. Once some success with Linux is established, that can be used to calm the fears of any doubters. It is important with new technology in a large company to have an early success. Even if the Linux project is simple, it gives you the proof that the technology works and the benefit of undisputable savings (for example, you did not have to buy another server). Once you have the proof that the technology works, you are ready to take the next step. StoreCompany started off with moving the firewall and proxy servers to Linux (see B.4, "Project 1: Firewall and proxy server"). That was a project of limited scope that proved successful.

The second reason for starting small is authority. The Linux-on-the-mainframe team needs the authority to move forward and to use the necessary resources. Big projects need significant resources, which again is often difficult to procure for as yet unproven technology in a company.

5.2.2 Initial server consolidation

For some, this level may be as far as they want to go. Here you might realize some hardware savings because you have consolidated servers on to the mainframe. This is also called simple server consolidation, taking the work from several servers and consolidating that work on one server. See Figure 5-1.

Figure 5-1. Simple server consolidation

With Linux on the mainframe, we continue to use the term server consolidation. It also means moving as much work as possible to one server. With Linux, what is often moved to the mainframe are whole images, one by one, including all the resources they need. Often this is done with the help of z/VM, in which case VM controls these guest systems.

Using Linux, this "work migration" is typically simple. There are no worries about how to pack all the things together, and usually the application we are talking about already exists on Linux. Thus, you do not need to migrate code from another operating system or talk your vendor into doing it. The only change that is made is where the image is hosted. Instead of on real, separate hardware, it is now on virtual hardware (LPAR and/or z/VM).

In this case, there frequently are savings on the hardware side, assuming that enough servers are consolidated. How many servers are "enough"? For example, one department store colocated 50 servers by going to Linux on the mainframe. The break-even point depends on your environment.

While what you include in a business case can vary from hardware and software costs to efficiency and ease of management, with Linux on the mainframe you can often make your case. What you have gained is a simplified setup, easy scalability, and improved availability.

Example

If there is already a mainframe present in the enterprise, then to achieve such a consolidation, the price of an Integrated Facility for Linux (IFL) should be considered. If a number of low-utilized servers, N, can be consolidated, that might make up the price for one IFL (see Figure 5-2).

Figure 5-2. Server consolidation cost example

Although this example is oversimplified, it illustrates why most of the larger consolidation cases you may hear about are on z/VM. With LPAR, you can have at the most 15 logical partitions in a z900 or z800, and with one server per partition, that is 15 servers. Under certain circumstances, even fewer images can pay off as a consolidation case.

What if you do not have a mainframe to start with? If you have to buy a whole new machine (say, a model z800 0LF that has one CPU), the cost will obviously be higher than adding one IFL to an existing mainframe. In this case, you need more Linux servers to consolidate for break-even. But the incremental growth path is attractive because if you need more capacity, then an additional CPU does not incur a great cost. By the way, the more Linux images you have, the more hardware sharing (such as sharing LAN adapters) will help save on cables and routers.

5.2.3 Full Linux integration in an enterprise

This second level of server consolidation is more complex because it involves integrating a Linux middle-tier with the legacy back-end systems. Because you have to use middleware and connectors and the like, this is a more complicated project. There are tuning issues as you allocate resources to the back-end and to the middle-tier. There are relative priorities that must be set to get the work done, and so on. Here IBM's WebSphere family of products can be effectively used for developing Linux applications and deploying them on the mainframe. For more information, see http://www.ibm.com/software/info1/websphere/index.jsp.