Linux on the mainframe — why it’s a good idea (an open letter to Sun)

On February 8th of this year, Sun announced plans to embrace the Linux operating system. Scott McNealy, dressed like a penguin, even said, “We love Linux, and I hope there isn’t any doubt about it.”

Apparently, you did not get the “memo,” as on February 20th, you issued a misguided condemnation of Linux and, more specifically, Linux on the Mainframe.

In sharp contract to Sun’s condemnation of Linux on the mainframe, the ever-growing enterprise Linux community feels that within seven (7) years Linux will become the dominant server OS in the datacenter, replacing UNIX-like systems such as Solaris, AIX, and HPUX. A unified Linux presence by Sun, Compaq, HP, Dell, and IBM will have enough momentum to effectively battle Microsoft’s entrée into the datacenter.

As Mr. McNealy previously stated, server companies love Linux. IBM loves Linux because according to Bill Zeitler the head of IBM’s server group, IBM invested $1B in Linux then “recouped most of it in the first year in sales of software and systems.[i]” At this rate of return, it’s easy to understand Sun’s fear of losing marketshare at the high-end and desire to reap the benefits of enterprise Linux.

Even the German government loves Linux, as their parliament recently switched from Microsoft to Linux, with one member saying it was irresponsible to entrust the work of
Parliament to “closed-source software.”

IBM’s recent announcement of the z800 Linux mainframe represents a big win for the Linux community. Linux is emerging as the default OS for major enterprise-class servers. Although Sun’s recent (and confusing) announcement about Linux tries to push customers to use Solaris at the core of the datacenter and Linux at the edge, the current trend indicates that Linux on the mainframe is used as the core of the datacenter and Linux on Intel-based servers are used as the edge.

While Sun’s Linux announcement of February 7 tries to position Solaris as the big brother to Linux, the Linux community feels that with the high reliability of IBM’s zSeries mainframe, Linux is positioned to dominate the core as well as the edge of the datacenter. In the statement, Ed Zander, Sun president and CEO, subjugates Linux to only “low-end servers.” However, the Linux community refuses to buy into claims by Zander that Linux will be pushed aside to the edge of the datacenter. The fact is that Linux is stable and scalable enough for the core of the datacenter.

On average, the mainframe has 50-year uptimes achieved through redundant processing unit (PU) error-checking, self-healing by automatic identification of damaged PUs and automatic replacement with spare healthy PUs embedded in the system, and the ability to hotswap hardware components with no downtime. When mainframe capacity is upgraded, system administrators can take advantage of the zSeries’s Capacity Upgrade on Demand feature which allows the switching on of additional CPUs, memory, disk and channels with zero service interruption.

Sun claims that Linux is “designed for Intel” and that “Linux on the mainframe is complicated.” This obviously shows a lack of understanding of Linux since Linux is widely available in Alpha, PowerPC, ARM, and Sparc systems. Sun must also know that some of Sun’s customers are running Linux on Sparc on partitions in Sun’s high-end servers to take advantage of Linux applications. Linux is appropriate for the mainframe and the mainframe is the standard for the core of the datacenter.

Further, the mainframe is specifically designed for high usage scenarios. For example, the processor cache memory has extremely high memory bandwidth. In the case of the zSeries, there are two 16 MB L2 caches that are shared by 6 to 10 PUs each. Unlike non-mainframe processor caches, the mainframe processor caches are highly connected with sophisticated interconnect logic. Again, the end-result is better performance from the hardware.

IBM’s VM, the most widely used and respected virtual machine technology in the world, has been refined over the last decade for high-end enterprise use and is actively used by the majority of Fortune 1,000 companies. Sun’s positioning of z/VM as a secondary layered OS is not accurate as z/VM has support for all the same hardware as z/OS and is used to run z/OS itself.

IBM is developing enterprise features that further integrate Linux and their mainframes including IBM’s Integrated Facility for Linux (IFL) which allows the management of multiple Linux images. This means less system administrators (and less cost) to manage Linux servers. Using LPAR and z/VM, mainframe users can run and manage multiple instances of z/OS and Linux on the same machine, getting the best of both worlds on a single server.

A few years ago, IBM announced the ability to run 41,800 instances of Linux on a single mainframe with VM and has since deployed Linux on VM at the majority of the Fortune 1000. Their recent announcement of their z800 Raptor Linux mainframe proves that there is serious demand in the enterprise for the Linux OS technology on the world’s most reliable hardware platform with the world’s most advanced and widely used VM technology.

While demand for Linux on mainframes remains high, Sun is correct that 20 Linux instances on the mainframe is not economically feasible for most users. However, the facts are again twisted. Sun claims low-end pricing on the mainframe starts at $400,000 ; however, according to IBM, pricing actually starts at $250,000. In addition, due to the high scalability of the mainframe, Sun’s example of consolidating 20 Linux servers is faulty. Mainframes are used to consolidate hundreds and, in some cases thousands, of Linux servers. With high numbers of users, mainframes have excellent ROIs.

For smaller server consolidation deploys, Linux, not Solaris, is also the best answer. Technologies such as Virtuozzo (www.virtuozzo.com) allow the consolidation of hundreds of Linux servers on commodity Intel-based servers costing less than $3,000. Although Sun claims that using Linux for server consolidation requires applications to be “recompiled and recertified,” the fact is that Linux server consolidation on Intel servers with Virtuozzo requires no recompilation or recertification. Solaris will encounter Linux not only on high-end mainframes, but also on lower-cost deployments with 4-way and 8-way Intel-based servers and even low-cost 1-way Intel servers.

Although Sun tries to position Linux on the mainframe as a closed and proprietary system, Sun’s technology remains more closed than Linux and is thus far less attractive to many system architects in enterprises. Although Sun releases some portions (not the interesting parts) of the source code for Solaris 8, the source code for the upcoming Solaris 9 remains closed. SWsoft went through the lengthy process of trying to get full access to the Solaris source code and finally gave up in frustration, forcing us to put software development for Solaris on hold.

No one can dispute that Sun is, and will always continue to be a Silicon Valley staple; however, what role your company plays in the future is not as certain. It is sure to play a larger role if it fully embraces Linux and dumps its current dual-OS strategy that is confusing customers. Linux will ultimately win on the edge and in the core of the datacenter. On behalf of the Linux community, which has grown to millions of people, companies and organizations around the world, I encourage you to recognize Linux’s place at all levels of the data center to work with the Open Source community to grow the server consolidation market.

If you like, we can send you a penguin suit of your own, or you can use Scott’s.

Regards,

Dr. Alexander Tormasov

Chief Scientist, Swsoft