10th Anniversary of Linux for the Mainframe: Beginning to Today
By Bill Claybrook
April 7, 2010
2010 marks the 10th anniversary of Linux for the mainframe. Here, Knowledge Center contributor Bill Claybrook delves into the 10-year history of Linux for the mainframe, discussing its first deployments, advantages and appropriate workloads, as well as its current market outlook, cost of ownership and available applications. He also offers advice on how you can determine if Linux for the mainframe is the right choice for your data center's server virtualization project.
The year was 1999. It was the beginning of Linux for the mainframe. IBM and SUSE (which was later acquired by Novell in 2004) began working on a version of Linux for the mainframe. By 2000, the first enterprise-ready, fully supported version was available: SUSE Linux Enterprise Server for S/390 [ http://www.novell.com/products/server/ ]. The first large, important customer was Telia, a Scandinavian telecommunications company. This year, 2010, is the 10th anniversary of Linux for the mainframe. The value propositions for Linux for the mainframe that were important in 2000 are still important today.
Linux for the mainframe began as two separate projects to port Linux to IBM mainframes. The first effort, the Bigfoot (i370) port, was initiated by Linas Vepstas in August 1998. Vepstas and his coworkers used IBM/370 mainframes at Princeton to do the port. A brief history of the Bigfoot project can be found here [ http://linas.org/linux/i370.html ] and here [ http://www.eweek.com/www.linas.org/linux/i370-bigfoot.html ]. The Bigfoot port became stagnant for political, social and market reasons after IBM announced the second port of Linux to mainframes (the Linux for S/390 port).
The Linux for S/390 port started in 1998 as a skunk works project at the IBM Boeblingen Lab in Germany, but IBM kept the Linux for S/390 port secret for more than a year. Speculation has it that secrecy was the order of the day because the developers feared political opposition from other parts of IBM. As a result, the S/390 team chose not to work with the Bigfoot team. The S/390 project was publicly announced on December 18, 1999. It was based on the 2.2.13 Linux kernel.
The first commercially available Linux for S/390 came from SUSE in Germany. SUSE began working with IBM, primarily at IBM Boeblingen Lab in Germany and Marist College in 1999. The Boeblingen Lab is only about a two-hour drive from SUSE's headquarters in Nuremberg. This explains the reason for SUSE's close relationship with IBM on the Linux for S/390 porting project. At the time of the Linux for S/390 port, SUSE had a number of different ports for Linux in-house, including ports for PowerPC, x86, Alpha, and SPARC.
In the 1999-2000 time frame, when SUSE was busy getting Linux up and running on IBM mainframes, there was no pressure (and no priority) to get involved with other mainframe vendors. Linux for the mainframe involved creating new processes and infrastructure at SUSE, a new business model, need for 24/7 worldwide support, synchronized IBM/SUSE Level 3 support processes, and ISV and IHV certification, etc. Interest by Fujitsu and Hitachi was more on the x86 and IA-64 (Itanium) platforms, even though their mainframes were compatible with IBM mainframes. Early versions of SUSE Linux Enterprise Server for S/390 ran on both Amdahl and Comparex machines, but the vast majority of mainframe customers in 2000 used IBM mainframes—and this is also the case today.
In order to get SUSE Linux Enterprise Server up and running on the S/390 mainframe, Marcus Kraft (the SUSE development manager for the port at the time) said that SUSE needed a compiler, library, and some base packages. The SUSE engineering team got most of this from Marist College in Poughkeepsie, NY. With the work that Marist College had done, the SUSE engineering team was able to get SUSE Linux Enterprise Server up and running on the S/390.
The next step was to gain access to a mainframe to continue the development and testing necessary to make the port of SUSE Linux Enterprise Server into a viable commercial product. SUSE got access to a machine from IBM and, with the aid of the SUSE Autobuild system, the company was able to get a system that worked up and running in about a week.
IBM had some legal problems around software drivers for Linux for S/390 because of the Intellectual Property (IP) in the drivers. As a result, the first version of SUSE Linux Enterprise Server for mainframes had object code-only drivers provided by IBM. Since then the IP problem has been resolved, and now all code is upstream and publicly available under the General Public License (GPL).
To get potential customers interested in Linux for the mainframe (in 2000, Linux on x86 was in its infancy) before the official release of SUSE Linux Enterprise Server for S/390, IBM and SUSE held a two-week-long install party. Because participants could not bring mainframes to the facility where the install party was being held, conference calls were used to walk interested companies through the installation steps. Many of the customers that participated in the initial install party are still loyal SUSE customers today.
First deployments on Linux for the mainframe
The first deployments on Linux for the mainframe were file and print servers. And the first piece of software that became popular was Samba. The first large commercial customer for SUSE Linux Enterprise Server for S/390 was Telia, the largest telecommunications company in Sweden.
The marketing literature for SUSE Linux Enterprise Server for S/390, published in 2000, indicated that the operating system supported IBM's G5 and G6 mainframe servers as well as the IBM Multiprise 3000. Shortly thereafter, support for the zSeries z900 in 32-bit mode was announced. SUSE Linux Enterprise Server for S/390 could run in a logical partition (LPAR), as a guest under the virtual machine operating system, and within the Virtual Image Facility (VIF) [ ftp://public.dhe.ibm.com/eserver/zseries/zos/vse/pdf3/wavv00/WAVV2000vif.pdf ].
In September 2000, IBM introduced the Integrated Facility for Linux (IFL) and the aforementioned VIF. An IFL is an IBM specialty processor dedicated to running the Linux operating system, with or without z/VM. The IFL was introduced by IBM to allow Linux operating systems and applications to run on the mainframe, freeing Linux on the mainframe users from paying the normally high cost of mainframe software licenses. An IFL is viewed as a single CPU for licensing purposes and microcode restricts IFLs to Linux workloads. IFLs are available for all IBM mainframes, as far back as the G5 series. Fujitsu and Hitachi also offer IFLs on certain models.
An IFL is not required to run Linux; Linux runs on general-purpose processors (CPs) as well. IFLs simply cost less. The initial price for an IFL was around $100,000 but IBM, in recent pricing changes, now sells an IFL for some mainframes (such as the z10 Business Class) for about $47,500. The price is about $75,000 on other larger mainframes. IBM says that approximately 4,600 IFLs were actively being used in 2008.
At the same time that IBM introduced IFLs, it also introduced a special, Linux-only, VM-like product, the S/390 Virtual Image Facility for Linux. It was aimed at IT staff previously unfamiliar with IBM mainframes. Shortly after the release of VIF, IBM found that z/VM was not difficult for IT staff to use and the S/390 VIF for Linux was discontinued in April 2002. IBM and SUSE laid the basis for Linux for the mainframe to be successful in 1999 and 2000.
Over the years, various features available on the mainframe have made their way into the Linux code base for multiple platforms, leading to significant improvements in the Linux operating system. For example, mainframe dynamic resource management capabilities have made their way into Linux for x86 platforms, and features such as the tickless timer have also made their way from the mainframe into the Linux code base.
Red Hat enters Linux for mainframe market
Prior to 2004, SUSE dominated the Linux for mainframe market, holding nearly 100 percent market share—primarily because SUSE was first to market with Linux for the mainframe and it had developed good rapport with IBM mainframe customers.
In 2004, Red Hat began paying some attention to Linux for System z servers, recording its first Red Hat Enterprise Linux for Mainframe sale. Red Hat's subscription revenue for Red Hat Enterprise Linux for Mainframe was less than $1 million in 2004. Also in 2004, Novell completed its acquisition of SUSE.
In 2006, Novell still had the Linux for mainframes market pretty much to itself, touting an 85 percent or higher market share. Red Hat was still focusing on the distributed Linux market and not paying much attention to Linux for the mainframe. After all, it was a less than a $20 million-per-year market for Linux subscriptions and the selling cycle was long. Novell also had an advantage over Red Hat because it got its mainframe hardware and software features into releases of SUSE Linux Enterprise Server for System z a few months before they appear in Red Hat Enterprise Linux for Mainframe.
In 2007, Red Hat began to take increased interest in Linux for the mainframe when it realized that there is drag-along revenue associated with each deployed Linux operating system on a mainframe. For every dollar spent for a Linux distribution, customers are spending two or more dollars for layered services, middleware, consulting, etc. In 2008, Red Hat's mainframe Linux market share jumped to 20 percent to 25 percent, with Novell's SUSE Linux Enterprise Server for System z holding 75 percent to 80 percent market share. The competition between Novell and Red Hat for the Linux for mainframe market was underway.
Red Hat has created a Fedora for System z project to help capture market share from Novell, while Novell continues to work closely with IBM customers, developing tools such as the SUSE Linux Enterprise Server Starter System for System z [ http://www.novell.com/partners/ibm/mainframe/starterpack.html ] that make it very easy for customers to try out Linux on the mainframe without spending little, if any, money.
Linux for the mainframe today
Mainframes are over 40 years old. Several years ago, mainframes were being written off as expensive, dated computers that were capable of running only large business applications. Not anymore. Mainframes have evolved rapidly during the past few years, with innovation from Linux and IBM.
Today, mainframes are inexpensive, faster and smaller than older mainframes. And, when used in server virtualization situations, they can provide significant savings on floor space and power. Take, for example, the IBM System z10 Business Class server [ http://www.ibm.com/press/us/en/pressrelease/25584.wss ]. The System z10 gives you equivalent capacity, an 83 percent smaller footprint, and up to 93 percent lower energy costs than 232 x86-based Linux servers. Cost per mips is one way that IBM compares the cost of mainframes.
Today, the cost per mips for the System z10 is a small fraction of the cost per mips for mainframes sold in 1980. This means that you get significantly more speed and capacity for the dollar now than you did with older IBM mainframes. You can get an IBM System z10 Business Class mainframe for around $100,000 and it gives you significant capacity with its new 4.4GHz, quad-core processors. Not bad when compared to the high prices of older mainframes.
Most of the server virtualization options are x86-based, and consolidating workloads onto mainframes [Herein, the word "mainframe" is synonymous with IBM S/390, IBM zSeries (z990, z890, z900, z800) and IBM System z9 and z10 servers] is often overlooked, but now it is becoming popular.
In many situations, it is the most cost-effective approach to server virtualization because mainframes aren't as expensive as they previously were, plus they have gotten much more powerful over the years. And you get the benefits of a mainframe such as increased hardware reliability, extreme I/O throughput, and less downtime. Those benefits just come with mainframes, and you won't automatically get them from server virtualization on x86 servers or any other architecture.
IBM's Rehosting Applications from Competitive Environments (RACEv) tool [ http://www.www.ibm.com/Search/?q=RACEv&v=16&en=utf&lang=en&cc=us&Search=Search ] is an excellent tool for evaluating your workloads, and to determine if they are a fit for Linux on the mainframe. The RACEv tool utilizes a set of workload categories, ranked according to how well they fit the mainframe. A RACEv evaluation tells you the cost savings you can expect to get by consolidating workloads onto Linux for the mainframe versus other server virtualization technologies such as VMware. It is a sophisticated tool that will give you a much-needed hand in helping you to decide what server virtualization technology is right for your data center, so you can save even more money for your company.
Owning a mainframe isn't costly
Mainframes used to be expensive. But not anymore when compared with old prices, the power you get, and how you can use them to save money, time and space in your data center. You can also save on Linux for the mainframe subscriptions from Novell and Red Hat. In 2008, IBM worked with Novell and Red Hat to create a pricing promotion for Linux for the mainframe distributions from the two companies.
Then, in late 2008, IBM created new pricing for the new System z10 Business Class mainframe to make it more economical to use. This amounted to about a 40 percent discount on Linux for System z subscriptions for the System z10 Business Class servers. IBM also dropped its prices for IFLs by more than 50 percent. These pricing promotions make server virtualization on System z servers even more attractive when compared to server virtualization on x86-based servers because it can be much more cost-effective on a mainframe.
To further improve your ROI through server virtualization and workload consolidation on a single mainframe, IBM (working with Novell and Red Hat) has created an IBM System z Solution Edition for Enterprise Linux [ http://www-03.ibm.com/systems/z/solutions/editions/linux.html ]. The Solution Edition is a packaged offering that brings together key components of hardware, software and maintenance at a low total cost of acquisition. It is built on the capabilities of z/VM and System z10 hardware. The Solution Edition targets smaller customers with stand-alone System z10 Business Class systems.
Save money on software licensing
Many large ISVs such as Oracle have per-processor licensing models for at least some of their software applications. This means you are charged for software based on the number of processors on the server on which the software runs. For example, Oracle Database Enterprise Edition is around $47,500 per processor. Generally, you pay that price regardless of the architecture.
Software running on multiple Linux VMs shares the cost of a single processor license. CPs and IFLs count as a single processor engine with respect to software licensing; that is, one IFL equals one core. Not only do you get a break on software licensing, but also the capacity in a System z10 Business Class machine is many times that of an x86 server.
Many applications available
Many applications are available on Linux on the mainframe. Many of the 3,500+ applications that run on Linux on x86 are available on Linux on the mainframe. More than 400 ISVs have certified applications running on Linux for the mainframe, including more than 280 IBM middleware applications. These applications range from data-intensive, high-I/O applications to CPU-intensive applications—including applications from BEA, CA, IBM, Oracle, SAP and Veritas. You can also run many of the important open-source applications such as Apache, MySQL and SAMBA on Linux for the mainframe.
With the new IBM System z10 server and its new quad-core processors (with its 70 percent more capacity and three times the available memory of the largest System z9 server), you can run CPU-intensive workloads that you couldn't run with any degree of efficiency on mainframes before. This greatly broadens the scope of applications.
Among the driving forces behind IBM's desire to get Linux on the mainframe was that IBM had made a decision that one operating system that could run on all of its platforms would be beneficial to IBM. SUSE had the technology, including Autobuild, to get the same SUSE Linux Enterprise Server code up and quickly running on IBM's iServer, pServer, xServer and zServer platforms. IBM's success caused other companies to follow suit.
Novell and Red Hat work closely with IBM to get new features incorporated into Linux for the mainframe. SUSE Linux Enterprise Server for System z and Red Hat Enterprise Linux for Mainframes are essentially the same operating system and are priced comparably. Both companies are using success with their large, x86-based Linux installations to entice customers to also use Linux for the mainframe.
To determine whether or not Linux on the mainframe is a good choice for your server virtualization project, take a look at how well the workloads you are considering consolidating fit the mainframe. The best-fitting applications are those that leverage the classic strengths of System z servers: high availability, high-I/O bandwidth capacity, etc.
Example workloads you might already be running include Oracle and SAP. Other applications that are good fits include DB2, Informix, WebSphere Application Server, and Apache. The breadth of good-fit applications has increased to include those with serious computational needs with the advent of the IBM System z10 server.
You should question some workloads for the mainframe. Applications that have not yet been ported to Linux, applications such as geological mapping animation rendering that are optimized for throughput, and applications that are too internally sensitive to try and migrate due to political issues are generally not good fits for Linux for the mainframe.
Bill Claybrook is a marketing research analyst with over 30 years of experience in the computer industry, with the last 10 years in Linux and Open Source. From 1999 to 2004, Bill was Research Director, Linux and Open Source, at the Aberdeen Group in Boston. He resigned his competitive analyst/Linux product marketing position at Novell in June 2009 after spending over four and half years engaging in cloud computing, software appliances, virtualization technologies, and numerous aspects of Linux platforms. He is President of New River Marketing Research [ http://newrivermarketingresearch.com/ ] in Concord, MA. He holds a Ph.D. in Computer Science. He can be reached at firstname.lastname@example.org.