Mainframe Migrations: Are MIPS the Right Measure?

When does migrating off the mainframe make sense, and should MIPS be the measurement standard we use?

Mainframe proponents like to say that Big Iron is more than holding its own, offsetting inevitable losses -- i.e., shops which determine to transition off of their mainframe -- with undeniable gains, chiefly in the form of bigger and brawnier MIPS deployments.

It's one of IBM Corp.'s talking points. Sure, IBMers like to say, some shops are transitioning off of Big Iron, but these tend to be mostly smaller shops -- say, organizations in which total mainframe capacity is under a certain MIPS threshold. Big shops -- where total mainframe capacity is above a certain MIPS threshold -- are either maintaining their MIPS investments or (in a number of cases) buying additional MIPS capacity.

IBM isn't the only one making this claim. Several mainframe ISVs -- including BMC Corp. and CA Inc. -- have adopted similar messaging at one time or another.

"If somebody is going to get off the mainframe, it's because … they have a very small footprint," confirms Robin Reddick, director of MSM Solutions Marketing with BMC. She cites BMC's recent mainframe survey, which found that roughly half of Big Iron shops anticipate adding capacity to accommodate new workloads. "We've found [that] the shops at 1,500 MIPS or lower tend to be more vulnerable to migrating off the mainframe."

Even asymmetrical mainframe rival Unisys Corp. -- which maintains two (still-converging) mainframe lines -- endorses this idea.

Cameron Jenkins, chief operating officer with Clerity Solutions Inc., a developer of mainframe modernization and migration offerings, begs to differ.

Sure, Jenkins concedes, there's a threshold at or above which shops are less inclined to consider transitioning off the mainframe, but there's still plenty of activity above this threshold, he says. More to the point, he argues, there's no technological reason that even very large customers -- shops of 10,000 or more MIPS -- couldn't migrate completely off the mainframe, either.

"We see most of [the migration efforts] between 100 MIPS to about 1,000 MIPS, although you [can] break that down further. Last year, for example, I would say the average was probably around 200 or 300 MIPS; this year, I've seen that go up to closer to 500, 600, or 700 MIPS in terms of the average," he comments -- in other words, basically in-line with projections from IBM, BMC, and others.

Booming Business

Jenkins and Clerity aren't exactly dispassionate observers, of course. While Clerity has a mainframe modernization practice -- in one prominent case, it helped Deutsche Rentenversicherung (DRV) transition a Big Iron-based pension management application from a z/OS to a zLinux context -- it also specializes in helping shops get off the mainframe once and for all.

Its bread-and-butter offering is UniKix, which -- although it can be used to support mainframe-to-zLinux transition efforts -- is usually used as a mainframe-to-Unix-Windows-or-Linux rehosting application. The bulk of Clerity's business, Jenkins points out, consists of mainframe-to-non-mainframe migration efforts. Right now, he says, there's little interest among Clerity's customers in transitioning traditional mainframe applications over to zLinux.

That being said, Jenkins doesn't think the mainframe is going anywhere anytime soon. Or ever, for that matter. In DRV's case, for example, rehosting on zLinux made sense, both computationally and economically. This will likely be the case for other large mainframe shops, too.

"I don't think mainframes are going to go away probably any time soon. That's like saying that COBOL is going to go away, and I've been hearing that for 20 years now," he explains.

"DRV did a move with us to the Linux partition. They moved [their] transaction processing workload to an [Integrated Facility for Linux] on the mainframe, and they continued to use the core mainframe for batch processing," Jenkins continues. "That just makes sense to them from a cost perspective; from a MIPS-utilization perspective, it was a good strategy."

If anything, Jenkins argues, CIOs are getting smarter about how they're using mainframe capacity. This, too, is something of a loaded claim. Although Jenkins and other Clerity officials are careful to stress that the mainframe has its place, they'll likewise argue that certain mainframe applications (e.g., CICS, IMS, or Adabase) can be run more cheaply in a non-legacy context.

Not surprisingly, Jenkins believes that "getting smarter" about the disposition of one's mainframe capacity often involves shifting CICS, IMS, Adabase, and other applications out of a z/OS context and into either zLinux (which keeps them on System z) or onto Linux, Unix, or Windows running on non-mainframe systems.

"Right now, I'd have to call [intra-mainframe migration] an outlier. With DRV, at one level, IBM was cooperating with us and helping us deliver a very successful project," he indicates. "Very high up, I'm sure that this [migration effort] wasn't something that pleased [IBM]. They aren't going to be going out and promoting that move as strongly as they could."

Not a Hard-and-Fast Threshold

Although Clerity sees a sub-1,000 MIPS sweet spot for mainframe migration, shops with ten times as much capacity aren't opposed to transitioning, either.

"A couple of years ago, we did what Gartner [then] said was the largest mainframe migration in North America; at the time, it was approximately 1,700 MIPS; now … the MIPS equivalent [of that migration] is now up to about 2,500 MIPS," Jenkins says. "Following this trend, we're starting to get requests that come in for 4,000 or 5,000 MIPS. We even had a request come in for 10,000 MIPS." Building out the equivalent of a 10,000-MIPS system in a non-mainframe context would be difficult, Jenkins concedes -- provided a customer wanted to do so in one shot. In reality, he says, such a project would evolve iteratively.

"I don't believe we're going to do those larger requests in one big bang. What I see us doing is actually breaking that down and doing it in stages; nobody just [allocates] 10,000 MIPS to one application," he points out.

The point, Jenkins concludes, is that an extremely large capacity footprint isn't always indicative of a difficult -- or prohibitively expensive -- migration effort. "I think that [a MIPS threshold] is largely a false number. For IBM to say the mission-critical applications are the ones that are running on mainframes or an LPAR region that's much larger than 1,000 MIPS, that isn't always true. We've seen equally complex [implementations] -- using [IBM's] definition of complex -- that are as small as 50 or 100 MIPS. These are projects that took us a year to complete, and they were only 50 MIPS. We've also done larger MIPS migrations that were [comparatively] very straightforward."