Is PowerLinux a New Platform? Not really...
jscheel 0600025BWM Comments (2) Visits (5340)
By: Jeff Scheel, PowerLinux Chief Engineer
A software provider recently said, point blank, at the beginning of our discussion, “Convince me that PowerLinux is not a new platform.”
The core of my answer went like this: the value of Linux is the same for software providers as it is for customers – Linux provides a single operating system environment across different hardware platforms. Customers and partners who have both x86 Linux skills and Power System skills have all the knowledge they need to run Linux on IBM Power Systems and PowerLinux servers. Customers and partners who only have x86 skills will need some training in PowerVM and the platform, but will quickly find commonality in the operating system.
This software provider's simple request strikes at the heart of the IT challenge – managing expense for the data center. Everyone is struggling to control software deployments in the enterprise. One can visualize this challenge as being depicted by a two-dimensional matrix (spreadsheet) of applications (across the top) against platforms (down the left side).
This complexity of this problem becomes most daunting when one considers two key factors:
The good news for all of us in the IT industry is that Linux and open source are driving commonality into software and redefining the “platform” to be the most basic components of the hardware platform. To fully understand this conclusion, we need to look at how the definition of “platform” has evolved over time.
Software Engineers typically draw the software stack with hardware on the bottom of the stack and applications on top, as shown below.
Applications run on operating systems, hosted by hypervisors, which virtualize the hardware architecture.
For the sake of simplicity, I have grouped several categories together. For example, middleware is really an enabling component of applications and as such could be a separate layer below “Applications.” Additionally, runtime environments like Java or Perl have been lumped into the operating systems level instead of being placed as an unique layer on top of “Operating Systems.” One could just as easily include hypervisors as part of the hardware architecture. To enable a technology trends discussion later in this blog entry, I have elected to explicitly separate “Hypervisors” as a distinct layer.
You must consider the visibility to users in the IT infrastructure of each software stack component to appreciate the impacts on the business. With a little insight, you can see that most people in the enterprise touch the application layer, while fewest touch the hardware. This observation suggests a roughly graduated pyramid, as depicted below.
The implication of this graduated visibility means that changing an application causes more expense for an enterprise than does changing hardware architectures, especially when changes in a top layer drive changes in the foundational layers below it.
Does anyone remember the day when applications ran on only one platform? Not too long ago, “platform” meant the whole software stack, a silo top-to-bottom. If I wanted to run the AmiPro word processor, I did so on my DOS-based personal computer. To do word processing at work, I used BookMaster on the mainframe. Applications were the “platform."
Over time, application providers began to write applications that could handle the disparate operating system environments. Applications (middleware) like IBM's DB2 database are available for Windows and UNIX operating systems. The advent of interpreted languages like Java further helped application portability. In this time frame, people began to think of the “platform” as the layers from the operating system downward.
With the adoption of the Linux operating system, the definition of “platform” continues to move further down the stack. When customers run RHEL 6.3 or SLES 11 SP3 in their environment, they get the same kernel version, built with the same compiler (which of course generated different binaries for the each processor instruction set), running the same libraries, leveraging the same file systems, and including the same levels of many common applications such as the Apache webserver. In this structure, properly written applications can expect common application environments, even though they may be running in different hypervisors on different hardware platforms. For well written applications, “porting” to a new platform generally becomes a recompile and a regression test.
The value of Linux to the enterprise comes from commonality. Differences drive expense. Commonality saves money. The greater the “visibility” for a component of the software stack, the larger the potential for savings from commonality. So, it should not surprise us that the historical trend of convergence has been top-down in our software stack.
While the majority of commonality has been with applications and operating systems, some convergence has occurred within the hardware architecture layer. I/O has developed standards like PCI. Networking converged to Ethernet, despite a competing standard of Token Ring. Likewise, storage area networking selected the Fibre-channel protocol over SSA. While these gains have been beneficial, they likely are limited in opportunity because processor architectures will likely never converge to a single architecture. Every processor type and system architecture focuses on solving a different challenge.
As we look forward toward the future, opportunity exists for continued convergence in the software stack with hypervisors. Today's enterprises have their virtualization options dictated by the hardware platform. Power systems virtualize with PowerVM, mainframes use z/VM and PR/SM, x86 systems use VMWare, HyperV, Xen, or KVM. Enterprises with heterogeneous hardware platforms run different hypervisors on each platform -- not by choice, but by mandate.
With Linux embracing KVM, the potential for a single, cross-platform hypervisor emerges. In the not too distant future, enterprises should be able to leverage KVM to reduce their virtualization expense, again changing the definition of “platform.” In this final picture, commonality has driven the definition of “platform” to simply being a new hardware architecture, reducing the impact (and resulting expense) to the most minimal definition.
Now that we have completed the analysis of the software stack convergence, we can provide a deeper answer to the original question: Is PowerLinux a new platform? Not really. Linux on Power is “just Linux”, RHEL 6.5 or SLES 11 SP3. For a software provider such as the one who posed the original challenge, their x86 Linux expertise combined with their Power System skills facilitate support of Linux on Power in a very cost effective fashion. As Power embraces KVM, even the hypervisor differences will be eliminated and the impact of adopting Linux on Power has truly been isolated to a few people for both software providers and customers alike.
While this convergence can be viewed from a historical perspective, one could also view these steps as phases of application maturity. New applications typically begin on a single operating system and hardware stack. Over time time, applications wishing to support multiple hardware platforms, will port. Modern applications, such as those written in interpreted languages like Java or Python or those compiled with using open source compilers like gcc, will migrate to new hardware platforms quickly. Linux and open source tools enable software vendors to maximize their addressable market by writing to common libraries and compiling using standard tools common to all platforms who run the Linux operating system.
The evolution of the software stack has been on an exciting trend. Open source software has driven and will continue to drive a convergence of the software stack as the IT industry evolves. We have come a long way from the days of “application silos.” Today's runtime environment with the Linux operating system provides a common architecture that enables cross-platform application support. With the emergence of KVM as a virtualization technology, convergence will continue into the hypervisor layer. Once this transformation is complete, platform differences will be reduced to the fundamentals of the processor architecture, enabling IT customers to minimize expense while leveraging the fundamental advantages each platform can provide to their solution. What enterprise would not want to be allowed the flexibility of selecting the best platform for their solution while minimizing expenses?