The current momentum behind IBM's S/390 and zSeries mainframes seems to be the most popular manifestation of a new trend in IT: server consolidation. After being viewed as legacy technology for much of the past decade, mainframes have suddenly become cutting edge again. However, while some IT professionals are championing these "big iron" systems as the cure to many current IT problems, others have their doubts.
In response to the recent article "Server consolidation: Is less actually more?" TechRepublic members took up this debate and contributed some important and timely observations. Their comments show that the issue essentially comes down to the matter of IT centralization vs. decentralization. So we divided most of the arguments into those that support decentralization (usually represented by Intel servers) and those that support centralization (the mainframe model).
Arguments for decentralization
- Phil.hall: "When those Big Iron machines crash, they take every service with them—not a happy situation. Better to think like a submarine and compartmentalize everything. I'd rather lose [just] the proxy than lose the proxy, gateway, mail server, and PDC. Like the old saying goes: 'Jack of all trades equals master of none.'"
- Wayne M.: "One issue that is not improved by reducing the number of servers is maintenance. The more users you put on one server, the more difficult it is to schedule downtime."
- Recoverer: "We're looking into consolidations now, and there are lots of issues. We've isolated applications on small dedicated servers for years, and we're running out of real estate. Pros for consolidation—more efficient use of space, less wear and tear on admins (and it is easier to manage two to three big servers rather than 20). Cons—reduced redundancy, higher cost, less flexibility. And the loss of flexibility may kill this initiative, since we have systems that don't play nice with each other. They all have unique processing schedules and resource needs."
Arguments for centralization
- Thorarinn: "I believe a Swedish telco has installed a zSeries running up to 10,000 virtual Linux servers for its clients. That's what this is good for. I also don't think there's any reason to worry about problems with scheduling maintenance or "losing everything" (proxy, e-mail, etc.) if it goes down. The 'z' in zSeries stands for (near) zero downtime. In other words, once you plug it in and start it up, you don't switch it off until you're going to throw it away."
- Jose P. Mir: "All systems will (sooner or later) crash. It's the IT department's responsibility to be prepared for that fact….My own experience tells me that those companies with 'micro' technology are more likely to suffer when the moment comes than those with 'real iron' in the core of its productive machinery. You can build an Intel-based LAN that offers the same MTBF and MTTR as normal mainframes, but it will cost you as much as the mainframe, and probably the performance will be highly variable and the maintenance cost will be higher!"
- Sujai.nath: "Everything crashes at one point or another, but the fact remains that those running 'big iron' such as our company aren't ready to junk them. Our mainframe has been up nonstop for over five years—show me a Windows cluster or even a UNIX cluster with that kind of uptime. These things are almost like people—on average, they have incredibly long lifetimes/uptimes. They can be 'operated' on while still working."
For more on this issue, member brant pointed to the Consulting Times' article "The Truth Behind the Great Server Heist," which provides an in-depth comparison of total cost of ownership (TCO) between a cluster of Intel servers running Windows and Exchange and an IBM mainframe running Linux and a comparable groupware application.
Finally, I thought one comment was especially poignant and should not be lost in all of this talk of mainframes vs. Intel servers. Therefore, I'll close with this comment:
RMoore357: "Every new management team has to make its bone doing something important. In IT, one of the fundamental games that managers play is centralized vs. decentralized. If the old environment is centralized, then the new managers claim that they can save money by decentralizing—and the same for the opposite position. By focusing on the things that help your case and ignoring/hiding the other factors, you justify your case. The reality is that you have to deal with both….Hire the best administrators that you can get—that will save you money—not some bright new machine."