Q&A: Rehosting Not Just a Pipedream
A Sun executive says mainframe rehosting is both affordable and practical
Tom Atwood is the director of business development for Sun Microsystems Inc.’s Enterprise Systems Products group. Atwood started out with supercomputer pioneer Cray Research, coming over to Sun when the Unix giant purchased Cray’s SPARC/Solaris business in 1996.
Atwood sat down recently with Enterprise Strategies to talk about Sun’s large SMP systems, the SunFire 12K and SunFire 15K, as well as about its mainframe and Unix rehosting initiatives.
Q. What’s driving demand for Sun’s very large SMP systems, such as the SunFire 12K and SunFire 15K?
A. If you look at sales of our [Enterprise 10,000, a predecessor to the SunFire 15K], probably about 75 or 80 percent of those customers had multiple domains [hardware partitions] in their systems; the average is three to five domains or more. You could make a very strong case that the majority of those customers were buying these systems to move over legacy applications and also to deploy new applications in a consolidated environment.
Q. Is much of the uptick of these systems driven by customers that are moving applications out of mainframe environments?
A. We have a mainframe rehosting group which basically handles this. There’s been a big demand for this [rehosting]. When you’re taking an older mainframe and you’re moving a couple of target applications off to Solaris, commonly you don’t need a lot of hardware. If you’ve got 1000 MIPs or smaller, you can get by with something like the 4800 or the v880. We have a customer, one of their target platforms [for a migration from a mainframe to Solaris] is a [SunFire] 15K, so this is a very, very large application. There are advantages to doing this, because going on a 15K allows you to keep the database, the manageability centralized, everything centralized.
Q. IT organizations are typically fervent endorsers of an if-it-ain’t-broke-don’t-fix-it philosophy. Why would they risk disrupting their operations to move applications that have functioned smoothly for decades to a new environment?
A. Well, they may not want to move all their applications. It may simply be more economical to move one or some applications over, particularly if the cost of running them on the mainframe is really expensive. But a lot of the time, the biggest driver isn’t wanting to migrate just legacy applications. Clearly, we have customers that buy systems with the intent to add more applications or grow [existing] applications. The nice thing about a larger server, and I don’t mean just the 15K, is that they have pretty significant headroom.
Q. What resources can you offer organizations that want to rehost their legacy applications on Solaris?
A. Well, like I said, we have a rehosting group, and because it sells the software and services, it tends to run a little bit autonomous. My expertise is with the overall ESP group, which includes [the mainframe rehosting] group, and they [the mainframe rehosting group] basically handle things like this.
Q. But you do offer some migration aides, such as a Mainframe Migration Toolkit, as well as a CICS translator and a batch manager for Solaris?
A. Yes, we do that. We at Sun take this very seriously, which is why we have a whole separate group that focuses on this. We have a Website [at http://www.unikix.com] that explains all about [our rehosting services]. There are whitepapers there [http://www.sun.com/datacenter/mainframe/whitepapers.html] that talk about some successful customer rehostings. There’s also more information about the resources that we provide and about other things that Sun provides for customers that choose to do this.
Q. Do you see a lot of Unix-to-Unix migrations? Customers moving from, say, AIX or HP-UX to Solaris? Or is it mostly smaller Solaris systems consolidating on large SPARC hardware?
A. We historically have seen both. Clearly, the easier migration is Sun to Sun, because you’re just going from one version of Solaris to another. Sun’s binary-compatibility strategy is very straightforward from 2.5 to 2.8 or 2.9. We have seen a lot of customers that have also moved from IBM or HP, so we’re getting both. Primarily, it’s many smaller servers to a few larger servers.
When you move from an HP or an IBM [to Solaris], there are some issues there. You may be moving from Oracle to Oracle, and that’s pretty straightforward, but we provide migration services for moving from AIX or HP-UX.
Q. Such as helping to port applications written specifically for AIX or HP-UX?
A. Our iForce centers are very valuable partners here. We’ve had partners that have been able to use iForce centers to do proof of concepts to show that certain types of applications can be ported, yes.
Q. You’ve said that consolidation scenarios, of legacy applications and existing applications alike, are driving uptick of large SPARC systems. What kind of partitioning strategy does Sun have to support consolidating all of these applications on a single large system?
A. Right now our higher-level partitioning is actually an electrical isolation, so I have to use the granularity of the hardware component, which is sold in either two CPU or four CPU arrangements. Most people tend to buy four CPUs.
Q. Mainframe environments boast very detailed granularity, down to the sub-processor level. What about customers who want to rehost legacy applications that need more granularity?
A. We have some people who would like to have greater granularity. Most people want [two- or four-CPU partitioning], I haven’t exactly seen why you want one CPU granularity on a single system, but there are people who want to run a single application on a single system. We have technology that allows us to partition our domains to a much finer granularity—this is a technology called Solaris Containers that we’re rolling out in a phased approach … [The first phase] is Solaris 9 Resource Manager, which used to be a layered application under Solaris 8. It’s now a kernel application. What Solaris 9 Resource Manager allows me to do is partition CPU resources in an individual instance of the operating system. If I want to run four applications in a single 4-CPU domain, I can use [the Solaris 9 Resource Manager] to assign shares [of the processors] to each application. Some may run well, some may grab too many resources. From a customer standpoint, they can try it and see which [applications] it works best for.
Q. Sun first shipped a 64-way system [the Enterprise 10000] back in early 1997. Since then, most of your competitors have caught up. Can we expect a larger system [than the 106-way SunFire 15K] in the future?
A. There is a limit to how big we can build the systems. When you’re talking about SMP systems, you have a tightly coupled interconnect. Every component has to couple with that interconnect. You have to keep it in a box, and there’s only a certain size that you can make that chassis and keep them in the computer. There is a limit basically on how big you want to build your box. I really can’t say that we’re going to build bigger or not, except to point out that there are limits. There are physical limits, too, to how big you want to make these things.
Q. What do you mean?
A. A lot of sites are pretty small offices, or under-sized. When I was with Cray, we sold one Japanese customer a CS-6400 [64-way system] that it ended up they couldn’t get in their elevator. So they had to get a crane and lift it up five stories and block off a lane of traffic in Tokyo. We had to think about things that it wasn’t designed for, like could we lean it on its side to bring it up, because it would be more stable that way! You have to look at things like that. Things that you wouldn’t always think about.
Stephen Swoyer is a Nashville, TN-based freelance journalist who writes about technology.