The Mainframe is No Dodo Bird

NEW YORK — Remember five years ago when Sun  and HP
 embarked on campaigns to disparage the mainframe?
Specifically, IBM’s  large refrigerator-sized zSeries
machines.

At the time, they had a market shift on their side; companies were fleeing
the comfort of the mainframe for smaller, more modular and cheaper Intel
servers. Rumor had it, they were easier to manage and no one could deny the
cost savings just from the price tag on the box.

But what many of these companies setting up farms of distributed servers
failed to recognize how much managing hundreds or thousands of servers would
drain the IT administrators’ time, or that the cost of powering so many
data-churning machines would be exorbitant.

With space, management and internal power cap mandates haunting many IT
shops, is it possible companies are coming back to the mainframe fold, which
so many server vendors and naysayers dismissed as a Jurassic period in the
computing world?

It’s quite possible from what I saw and heard at an IBM System z Summit
mainframe glorification event
here this week.

Steve Mills, senior vice president of IBM Software, said
space considerations, cooling and other factors are leading businesses to
look at how they bring workloads together. And they are looking to
mainframes for this consolidation.

“Does it matter what the physical box looks like, what the operating system
is? Are these religious issues important issues that need to be debated
today? Or are the important issues that need to be debated today how to get
the more efficient execution of my business… and free up dollars to
reinvest in new applications and new capabilities?

“All of that is leading to very fundamental and profound shifts in the
attitudes and buying habits of businesses around the world. [Customers] are
looking increasingly at consolidated computing choices, and clearly the IBM
zSeries is the premier consolidated computing platform in the world today.”

Nice pitch, Steve.

Jim Stallings, general manager of System z for IBM Systems & Technology
Group, chimed in with some stats to bolster Mills’ proclamation.

Mainframe sales and processing units, it seems, are growing. In IBM’s fourth
quarter 2006, System z sales were up 12 percent year-over-year, while MIPS
— the way a computer’s speed and power are measured —
grew 9 percent to tally over 11.3 million MIPS sold. In fact, Stallings
said, IBM shipped more MIPs in that quarter than it did in all of 1997.

Twenty-five percent of the total annual MIPS IBM ships are based on Linux,
while 60 percent of the MIPS shipped are a combination of Java and Linux.

Moreover, IBM’s specialty mainframe configurations for Linux (IFL),
information integration (zIIP) and application performance (zAAP) also sold
well, boasting MIPs tallies of 1,200, 1,500 and 1,700, respectively.

Stallings said that after traveling the world and talking to customers about
their pain points, he is convinced customers are rationalizing their IT
strategies in a different way so it consumes less power, reduces complexity,
and scales up without losing value.

“Power in many cases is out of control,” Stallings said, displaying a chart
from researcher IDC that said the power associated with running the server
will cost twice what the server costs two years from now.

“The acquisition cost is no longer the driver of the conversation. It’s,
‘What does it cost to manage it?’ The biggest cost to manage it is people. The
next biggest cost is power. It’s given rise to the mainframe being in more
conversations and more deployments and more services engagements than ever
before.”

In short, he said customers are looking to virtualize as much as possible.
Thanks to the increase in virtualized memory for the IBM System z’s z/VM
virtualization software, the ability to move many servers onto one mainframe
became more definitive. Moving hundreds of underutilized Intel x86 boxes to
a few larger machines may prove more attractive to some enterprises.

Next page: Hoplon Infotainment gives it a go

Page 2 of 2

Hoplon Infotainment is a case in point.

The startup, which proposes to offer massively parallel online gaming for consumers, ran out of processing room on smaller commodity boxes. Each box could only accommodate 20,000 users, and users hosted on separate servers could not play with users hosted on other servers, creating a wall of separation.

Hoplon CEO Tarquinio Teles said his company leased an IBM System z mainframe and solved the problem of allowing thousands of gamers to interact online.

But don’t take it from Mills, Stallings and an obviously enamored IBM business partner in Hoplon.

Analysts are beginning to see IBM mainframe uptake, too. I asked WinterGreen Research analyst Susan Eustis if she’s encountered customers moving away from Intel boxes to mainframes, and more importantly, why they are doing so.

“I talk to a ton of customers about that,” Eustis said. “If you take 13 servers to power one application, you might have four guys on it — two technicians and two developers.”

“Whereas if that application is virtualized on a mainframe, it’s probably taking 13 MIPs and some small percentage of one person’s time, so your labor costs, your management costs and your power costs. For the mainframe, your power cost is a very, very small percentage.”

Eustis added that while datacenters running several distributed servers may cost $60 million, the cost of the mainframe to handle the same application is $6 million. Net-net, a 10-to-1 cost advantage for the mainframe as a shared workload environment is not a bad return.

Fine. So you save some serious bucks using a mainframe. I’ll buy that for now. How is the mainframe’s reliability? Eustis said the mainframes she’s inquired about suffered about five minutes of downtime per year while no one using servers has been able to get above three-nines availability.

So, if mainframes are so great, why did some shops junk them for commodity Intel boxes? Eustis has an answer for that, too.

“I don’t think they did a cost analysis,” Eustis said. “It’s very compelling if you’re running a department to run out and buy a few servers to get your application up and running. But all of these companies added more and more Web applications, and that’s how you had this explosion of cost that no one ever looked at.

“Now, the enterprises who are able to control cost and look at their cost are systematically saying ‘the mainframe is 10 times cheaper.'”

If there is one cautionary tale to switching to the mainframe, Eustis said, it’s control.

“The thing that IBM is not talking about here is if you move all of these applications back onto the mainframe, will the departments lose control of their applications in a service-oriented architecture?”

Fair enough, but with the dollar savings, it’s no shock why cost-conscious businesses are moving to the mainframe. Maybe that explains why Sun and HP have toned down the anti-mainframe rhetoric.

Clint Boulton is managing editor of internetnews.com.