In-Depth

Big Iron and Open Source Join Forces

IBM Adopts Linux

• By Jon Udell
• 02/01/2002

Under IBM's venerable wing, Linux—and the universe of software that comes with it—has become a fully supported mainframe operating system. "We are so confident of how reliable Linux is," says Dan Frye, director of IBM's Linux Technology Center, "that we provide service and support contracts that guarantee the same levels of response time as for our commercial operating systems."

This unlikely alliance between two very different computing cultures creates an immediate short-term opportunity to consolidate PC server farms, saving electricity and floor space as well as software and administrative costs. More broadly, as Web services become the means of cross-enterprise application integration, Linux with local access to mainframe applications and data represents a powerful integrator's toolkit. Along with Java, which was already supported natively on the mainframe, Linux brings a panoply of open source programming resources that speak the protocols of Web-services-based integration: XML, SOAP (Simple Object Access Protocol), and WSDL (Web Services Discovery Language).

Of course, Linux doesn't run on the zSeries mainframe only. It's supported right down the line, on the iSeries (AS/400), pSeries (RS/6000), and xSeries (Intel). This familywide Linux embrace was the key to IBM's Linux strategy, according to James Barry, now CTO of Jabber Inc. Formerly an IBM program director, Barry wrote the business plan for IBM's WebSphere application server. "There were 15 different app servers," he recalls, "but many different OSes were in play, and there was nothing that could run across the whole scalable family of servers." He pushed for Linux as a single unifying standard. What helped enormously was a groundswell of interest within IBM. Internal mailing lists about Linux were flooded with traffic. "If you said you wanted to do a Linux port, you'd have programmers lining up," says Barry. "When you want to get the bright minds, you get them with Linux."

The cultural cross-fertilization flows in both directions. At Infocrossing, a managed service provider in Leonia, N.J., Linux running on an S/390 is the preferred platform, favored over Windows/Intel or Solaris/SPARC. With many years of mainframe experience under their belts, Infocrossing's engineers have long since internalized data-center management skills and traditions that server-farm operators are now learning anew, according to Tom Laudati, Infocrossing's senior VP for enterprise engineering. "We think about backup and configuration management in a whole different way," he says.

But Linux's open source culture can also pleasantly surprise the mainframe crowd. Rich Smrcina, a senior systems programmer for Sytek Services, is a mainframe consultant who knew nothing about Linux until two years ago. At that time, his company undertook a pilot program to explore mainframe-based server consolidation under Linux. The experiment succeeded and now one of Sytek Services' customers runs a suite of Linux-based services—including DNS, mail, network monitoring, and a DB2 connector—on a mainframe under the VM hypervisor.

As Smrcina ported other Linux apps to this environment, he learned two things. First, porting wasn't an issue. Linux on the mainframe is just like Linux everywhere else, and applications like Samba, Apache and INN just compile and run "right off the wire." Second, the culture of open source development has its own rich traditions and long history of quality results. "Because nobody owns the code," says Smrcina, "there can be a perception that these are just hackers with no idea of how enterprise software should work." That couldn't be further from the truth, he says. "I've seen that this code is tighter and better than many commercial offerings. Everything is peer-reviewed, contributions are accepted only from a select few, and everything goes through rigorous testing."

Nowadays, some of those peers are IBM employees who themselves also contribute to Linux. In Dan Frye's group alone, there are 250 programmers working to beef up Linux's enterprise functionality. He ticks off the projects: Volume management, file systems, networking, security, systems management. The scope of IBM's involvement in Linux, Smrcina says, "has blown me away."

The Computer is the Network
The mantra for the era of distributed Web services has been Sun's "The network is the computer." One pundit, writing about IBM's evolving mainframe strategy, quipped that IBM should propose the inverse: "The computer is the network." It's long been true that virtual servers on the mainframe have been able to form an internal TCP/IP network, communicating faster and more reliably than over a LAN. The advent of mainframe Linux dramatically expands the range of software that can exploit this capability. For Smrcina, this meant moving a DB2 connector from a LAN-attached NT box to a Linux image on the mainframe. In addition to consolidating an external server into the mainframe, the Linux-based version of the DB2 connector eliminated a network hop.

A new technology available with IBM's z900, called Hipersockets, turbocharges that internal LAN. Conventionally, although traffic on internal TCP/IP connections never leaves the mainframe, these connections still use the network subsystem, OSA Express. With Hipersockets, connections use direct in-memory data transfer. "This is full TCP/IP support," says z/OS product manager Mary Moore, "so protocol overhead means you don't get the full speed of native memory." Still, there's no faster way to sling massive datasets across partitions and virtual servers. Because Hipersockets drivers exist for Linux, applications running in Linux images can enjoy this hotrod method of data access. Harry Roberts, CIO of Boscov's department stores, says that "Hipersockets will allow us to collapse all our DB2 resources onto one LPAR [logical partition]."

It's great to minimize communications overhead, but issues of reliability and manageability tend to loom larger in conventional distributed systems. A typical PC-based e-mail system comprises a group of servers in central and branch locations. The problem, according to Greg Olson, cofounder and CEO of Sendmail.com, is that you can't deliver mainframe-class reliability to all those locations. Sendmail.com offers a commercially supported version of the software that powers many of the world's e-mail servers. [See "Rain, Snow, Sleet or Spam" on page 52.—Ed.] The mainframe, he says, is a "phenomenally good" platform for this software. "The MTBF for a mainframe is 60 years," he points out. "So you don't have to say, oh well, Cleveland is down at the moment. You can have everything working all the time."

E-mail is growing at 270 percent a year, and is increasingly mission-critical. Olson says that the mainframe has the capacity to handle these loads. The only reason to distribute e-mail service, he says, is when limited bandwidth to a location mandates pushing the service to the edge of the network.

As external workloads such as e-mail service consolidate into the mainframe, it's vital to ensure adequate levels of isolation and quality of service. These, of course, are just the problems that mainframes are designed to solve. It's easy to set quotas on the CPU and I/O resources allocated to a processor or partition, and this, says Olson, "makes it safe to consolidate e-mail onto your accounting server." Conversely, it's straightforward to give more resources to an application that's experiencing peak load. For Tom Laudati at Infocrossing, that means the difference between over-committing PC or SPARC boxes that do nothing most of the time, and simply reallocating mainframe resources. Traditionally, such adjustments required human intervention. But now, says Mary Moore, IBM's Intelligent Resource Director can increasingly automate the task. "It takes your service agreement, written almost in English," she says, "and behaves like a performance analyst working at hardware speed, floating cycles and channels across LPARs."

The Cost of Ownership
At Boscov's, Harry Roberts is consolidating 70 NT-based servers onto a set of Linux images running under VM. The software mix includes invoice processing, price management, a gift registry, and other two- and three-tier applications talking to DB2. While the NT server farm is scalable, he says, "the care and feeding costs are exorbitant." One reason is headcount. He has to dedicate one administrator for every 10 to 12 NT servers. The other is the rising cost of Windows ownership. "As soon as you move en masse to Windows 2000, XP comes along," he says. "This didn't look like a prudent direction for us."

Bill Carico, CEO of ACTS Corporation of Kingsland, Texas, was among the earliest to deploy an application for Linux on the mainframe. The application, TestManager, is hosted at Infocrossing, and conducts exams over the Internet. Users, interacting with the software through a browser, have no idea there's IBM big iron on the other end. And that's just how Carico thinks it should be. It means that a small company like ACTS can go to market as an ASP, and can afford to deliver Web-based software with mainframe quality of service. What's more, Linux brings formerly prohibitive mainframe software costs within reach, "freeing you from the shackles of the ISV bandits."

Linux, however, brings its own costs of ownership. For normal mainframe administrative chores, such as workload and volume management, the traditional tools and techniques work the same way for Linux as for any other guest OS. But Carico notes that mainframe pros don't make instant Linux sysadmins. "It's easy for a senior OS/390 guy to spend all day chasing down some little Linux quirk." On the other hand, college graduates increasingly show up for work with Linux proficiency. "Put Linux on that mainframe," says Jabber's James Barry, "and you'll attract new blood."

According to Mary Moore, IBM is working on new management tools that will make it easier to deal with sets of cloned Linux images. Historically, z/VM didn't run lots of identical guests, but now that people are seeing a quick return on mainframe-based Linux farms, she says, "we're concentrating on ways to streamline and manage replication of Linux images."

The Linux Software Universe
When there's mainframe capacity to spare, dedicating a fraction of it to Linux looks like a great solution to a certain class of problem. Today, according to Pierre Fricke, a D.H. Brown analyst, Linux qualifies as an enterprise infrastructure system. When mainframe capacity is available, this may be the most cost-effective and reliable way to deliver Internet services. But it isn't yet what Fricke would classify as an enterprise application system. A number of key line-of-business apps from the likes of Oracle and PeopleSoft aren't yet available for Linux. And some that are haven't yet been certified for Linux on the mainframe. "Just because an ISV supports Linux," says Fricke, "that doesn't mean it supports Linux/390." Infocrossing's Tom Laudati agrees that ISVs aren't yet flocking to the platform.

But while Fricke thinks that bread-and-butter enterprise applications "aren't coming to Linux/390 anytime soon," Harry Roberts believes the opposite is true. Almost all of his major vendors have announced they intend to port to Linux. And when these applications become available, he plans to move over to them.

Even if Linux is only providing enterprise infrastructure at the moment, that's nothing to sneeze at. "Linux brings the Internet right onto the mainframe," says Sendmail.com's Greg Olson. That means, he explains, that Linux can exploit the mainframe's strength as a reliable and high-capacity platform, while creating an integration space in which to Internet-enable legacy applications and data. Fricke agrees. As the Web services movement progresses, he says, Linux's rich software toolkit makes "a flexible and secure exposure mechanism for proprietary apps."

For mainframe pros and Linux mavens alike, the union of these two cultures brings extraordinary opportunity. The Web is vast, and Web-enabled businesses face scale-up challenges beyond anything formerly imagined. Traditionally, there were two ways to attack the problem. You could get big, or you could get distributed. With Linux on the mainframe, IBM's message is that you don't have to choose. You can get very big, and enjoy all the benefits of bigness: High-speed internal communication, security, ease of management, optimal resource utilization. At the same time, you can stay distributed—that is, flexible, modular, portable, interoperable.