In Part 2, Sun responds that overall system architecture, and not just energy costs savings, should determine which server platform to use in the glass house, while author Scott Courtney does an in-depth comparision of the two glass house platforms--a mainframe running Linux and a Sun server farm running Solaris.
As you might expect, Sun Microsystems Inc. sees such comparisons from a very different perspective. For starters, the comparison between an IBM zSeries mainframe and Sun's UE250 machine (one of their smaller servers) is intrinsically an apples-to-oranges situation. More than just a disparity in size, the fundamental architecture of the machines is vastly different.
Chris Kruell, group manager of Sun's System Products Group, responded this way: "Why isn't IBM comparing their mainframe server to the Sun Enterprise 10000 server? Or comparing an array of Sun Enterprise 250 servers to an array of IBM B50s?"
Rich Ptak, speaking for The Hurwitz Group of Framingham, Mass., which did the energy cost comparison, defends his company's decision to compare different systems because the choice between these two disparate options is precisely what their consulting clients must evaluate. "The object," says Ptak, "was not to compare [zSeries] against Sun's offering in the mainframe space, but to compare against a client/server configuration." Sun was chosen because of the company's prominence in the distributed server marketplace.
Sun Microsystems offers a machine called the Sun Enterprise 10000 that is in the same league as the IBM mainframes. It even offers a capability called Dynamic System Domains that is analogous to IBM's LPAR (Logical PARtition) capability in the S/390 and zSeries machines, and which allows multiple operating system instances to share the same physical hardware in a dynamically-alterable way.
So although The Hurwitz Group didn't compare against the Sun E10K in this particular study, Rich Ptak says they could indeed do so in the future. The purpose of this study was to compare horizontal scalability versus vertical scalability, and Ptak says "it's certainly possible" that the E10K could offer many of the same advantages as the IBM mainframe in terms of conserving power and floor space in a data center.
Sun's Chris Kruell also asserts that cost savings aren't the only issue that should be considered. His argument? "The pennies [a customer] would save in electricity do not make up for the extremely limited functionality available in Linux on s/390." IBM's LPAR design, he says, requires customers to "commit to an expensive, proprietary technology to get flexibility and manageability of having multiple domains within a single server." Kruell says Sun's Dynamic System Domains offer full isolation between operating system instances, just as IBM's LPARs do. "A domain is completely shielded from any type of software error from other domains, including those errors generated by OS panic condition or an application program crash. Also, hardware errors contained within a single Dynamic System Domain will not affect another domain. If memory or processors fail within a domain, any component specific to one domain cannot affect any other domain." Up to sixteen DSDs can be defined per physical machine, a number which is very comparable to IBM's LPAR capabilities but which is considerably smaller than the hundreds or thousands supported in IBM's Virtual Machine (VM) environment.
Linux on zSeries or Solaris on Enterprise 10000?
Of course, Sun's servers don't run Linux and IBM's do. IBM supports Linux because of its adherence to open standards and its portability across their widely divergent hardware platforms. Sun, which enjoys a tighter level of binary compatibility across different platform levels, sees Linux as an operating system that is not yet as robust as Solaris. Kruell refers to the S/390 port of Linux as being in an "early experimental stage." IBM, he says, wants to use Linux as "a means of stringing customers along with the hope of a new operating system that will revive a dying platform, thus preventing customers from migrating their strategically important applications off of the proprietary mainframe and onto more open UNIX systems."
"At this point," he adds, "enterprise-class Linux is not ready for the data center, and it will be several years, at best, until this changes."
This doesn't mean that Sun opposes open source tools or applications. In fact, the company has a Linux Program Office headed by Herb Hinstorff. Hinstorff emphasizes that Sun is already shipping Open Source tools as a standard part of their Solaris packaging, and that Sun is committed to working cooperatively with selected third party and Open Source initiatives. "We're working," he says, "on a variety of projects including contributing engineering time and code to ensure that both the Linux and the Solaris Operating Environment (SOE) are drawn more closely together. Examples include the University of Michigan, contributions to Apache and contributions to the GNOME area. We provide key Sun technology for Linux including: Java J2SE, J2EE, Forte for Java Toolset."
Two Viewpoints, One Question, Many Answers
IBM and Sun clearly view the economics of data centers very differently, and both companies make persuasive arguments that their offerings are a better choice. The tempting question to ask is, "which one is the best?" Answering this simple question, though, may not be a realistic goal unless it is considered in the context of a specific enterprise's needs.
There are a few points that seem to be generally accepted by both companies:
The savings in electricity and floor space from consolidating racks of smaller servers into one large machine can be substantial, whether that large machine is an IBM zSeries or a Sun Enterprise 10000. In addition to the direct cost reductions, there are efficiencies in system administration and management that result from physical consolidation. There are benefits from the load averaging of bursty applications that bring more efficient utilization of CPU capacity versus physically isolated machines. And while there appear to be security concerns from some hosting customers with regard to sharing hardware platforms with other customers, both IBM and Sun seem to feel that their solutions sufficiently isolate applications so that these concerns are not a problem in reality.
IBM sees an across-the-board Linux adoption as critical for integrating their disparate platforms, from top to bottom. Sun, with the luxury of binary compatibility, sees open source tools and Internet standards as being more important than Linux itself. Both companies have political agendas at stake: IBM needs to leverage their prominence in the legacy data center, while Sun needs to retain the mindshare of large Internet providers and other traditional UNIX customers. Both companies want to sell hardware, middleware software, and services.
Sun has an edge in terms of fixed-cost barriers to entry. The "cover charge" at the Mainframe Bar and Grille is steep compared to the Solaris Cafe. On the other hand, IBM's VM operating system, while pricey, can do things that have no parallel on Sun hardware. Only IBM can run hundreds or thousands of operating system instances, simultaneously. Only IBM can run a UNIX-like operating system right alongside the legacy business and financial systems still contain the vast majority of corporate informational assets. IBM says this is an incomparable advantage; Sun says it is irrelevant because legacy systems can be linked to e-business using Internet protocol standards.
In the final analysis, every business has different needs and those who would offer application hosting to multiple clients must evaluate and compare a variety of conceptual approaches. One thing that is certain, though, is that companies have more choices today than they did a few years ago. Competition spurs innovation, and the customer reaps the benefits.
Scott Courtney is technical editor at internet.com.