Q&A: Kicking the Tires of Mainframe Linux
The allure of Linux is strong, but unless ISVs rein in their software prices, Linux's growth on the mainframe could be limited
Mark Post is a senior infrastructure specialist in Electronic Data Systems’ (EDS) Software Services Group. He’s also a prominent participant in the mainframe Linux community—he lectures at SHARE and other events, and is Webmaster of the linuxvm.org site. Post has been involved with Big Iron Linux almost since its inception.
As author of an official IBM Redbook (“Linux for zSeries and S/390: Distributions, SG24-6264”), Post currently heads a group at EDS charged with supporting Linux in the Americas. We talked with him about EDS’ portfolio of Linux services; the allure of Big Iron Linux for non-traditional applications; why Big Iron Linux is still an under-funded, quasi-exploratory effort in many organizations; and what IBM and other vendors must do to spur further uptake of Linux on the mainframe.
Big Blue has done a great deal to make the mainframe a more attractive target for Linux deployments, Post agrees, but unless ISVs rein in their software prices, uptake could still be limited.
How did you first get involved with Linux on the mainframe? Was it because of something that you were doing for EDS, or was it the result of a side project or something similar?
I’d gotten involved with Linux outside of work a couple of years prior to getting involved with mainframe Linux and had started to learn a lot about it. I played with IBM’s Unix System Services, so I’d gotten to know a little bit about Unix and decided I wanted to learn more about it in a play environment rather than a production environment. So I borrowed a Linux CD set from a guy at work and took it at home and bought a book and started learning as much as I could. As I compared the two, I became increasingly frustrated with what I perceived as the inflexibility and incompatibility of Unix System Services (USS), and I thought, “Wouldn’t it be great if there was a version of Linux that ran on the mainframe?”
This was May of 2000, just a little bit after the patches [for Linux on S/390] had been announced on the Internet, when I found out it would run under VM. I’d supported VM in the past, and I could immediately see the possibility of running lots of these [Linux images] on the mainframe, and not stacking up the hundreds of servers EDS had been buying.
What kinds of services does EDS offer customers around Big Iron Linux?
In the mainframe environment, the official [EDS] service offering is still under development, so that’s a little more fragmented … so the result is that right now, [when customers approach EDS about mainframe Linux] it’s pretty much a custom effort every time. We have clients in North America as well as Asia-Pacific that are just really starting to get a toe in the water, so to speak, in terms of checking out which applications run well and which ones don’t, which ones they’re willing to take a risk on early.
Some of our larger clients are familiar with Linux perhaps, but not on the mainframe, so they’re being fairly cautious. They’re not really hungry for it yet, although I think that once we have an offering, that will change.
Are there specific kinds of customers—that is, they’re in specific verticals, they have certain requirements, and so on—that are drawn to Big Iron Linux?
From what I can tell so far, they’re clients that already have some work on the mainframe, so they understand what that platform is capable of in terms of reliability and usability. When we’ve broached the subject with clients that are not running any mainframe workload, they’re really unsure. For most of those clients, all they’ve ever heard are that the mainframe’s a dinosaur, the mainframe’s dead, and so on… So the ones I’ve seen that do have the mainframe workloads and are interested in mainframe Linux, they come from a lot of different industries. I can’t say that they’re any particular ones, although most [customers] do have very different requirements.
One client was interested in making sure that mainframe Linux could interoperate in a grid, for instance, and so we got a proof of concept set up for that, and they were satisfied with that and they went to look at particular apps that would be transferred from their existing platform to mainframe Linux.
On other platforms, Linux has been a very popular choice for FTP, Web, and SMTP servers. Is that where a lot of the Big Iron uptake is, too?
They’re not terribly interested in doing FTP servers, Web servers, [and] DNS servers, although I certainly think that’s where they can get a lot of bang for their buck. The easy stuff to do is this infrastructure stuff, SMTP, Web servers, etc., but it appears that in most of the cases I’ve seen, they’re interested in doing more than that, and in one case, we told a customer that they’re being a little too aggressive in terms of what they want to do. They wanted to go with something really big and critical upfront, and we had to say, you don’t know how any of this stuff is going to run, let alone the critical stuff.
Why do you think customers want to deploy what sound like unusual or pretty critical workloads on mainframe Linux?
First of all, what I see based upon the people that I talk to at SHARE and places like that is that most of the systems type people that wind up getting involved with this just in terms of sheer numbers are MVS system programmers, and part of that is that there are just more MVS installations out there than anything else. They’re mostly told to do this because it runs on the mainframe, and they start out feeling totally lost, because it’s just outside their experience. A lot of other people recommend really, really working with the midrange folks who have been doing stuff like this for years, if not let them do it completely. Unfortunately, a lot of people don’t have that flexibility, but the ones that do wind up getting a smoother uptake on this than when they’re just doing it on their own.
So a lot of them aren’t necessarily doing this on their own, they’ve got a directive from management?
Absolutely. When I give my talks [at SHARE and other speaking engagements], I take a poll at the beginning and ask how many [attendees] are there because [they use] MVS, VM, or Linux—we actually do have some shops that are … thinking about primarily putting Linux on mainframes, even if they’ve never done anything with mainframes before—and how many people are there because they’ve been told to, and a lot of people raise their hands. In some cases, they were just literally ordered to be there, and at least see if [Big Iron Linux] was something that their shop needed to look at or not. They knew that they needed to make that level of judgment about it before it went for too much longer and they were hopelessly behind the curve They’ve basically been ordered to do this. They’ve just been told “Do this, and don’t spend any money, we’re not going to send you any money.”
When they decide that Big Iron Linux is something that they want to do, is that typically when they decide that they’re in over their heads and that they need the expertise of someone like EDS?
Even once they decide to get serious about it, they really, really want to keep their costs down, so they tend more to dump it on their existing folks and have them struggle, rather than come to either IBM or EDS in terms of expertise. For these people [who’ve had a mainframe Linux project dumped in their laps], there’s a lot of help out there in the form of the [Linux for 390] Marist mailing list, and if they do come to SHARE, they come to our session where we’re giving them help there. But unless there’s a real commitment from management, they’re not coming to us.
You’ve written an official Redbook for IBM about the different Linux distributions available for mainframe Linux. What are the most popular distributions, and are some better than others for specific purposes?
Generally speaking, it has been SuSE that’s been predominant, because they were first in the market with a commercial distribution that had support associated with it. There have been others before it, but they weren’t really by well-known organizations. They weren’t really supported, sometimes not at all. Even now, it’s SuSE by a long-shot, but now that Red Hat has their Enterprise Version 3 out, I’m starting to see in terms of new installations, much more even number of Red Hat trials versus SuSE, and there have been some defections back and forth.
Speaking of defecting, is it relatively painless for users to move applications from one mainframe Linux distribution to another?
There are differences in what ISVs applications are certified for which distributions. There are differences in the glibc and share library environments on the two, so you can’t just move your application. You’re probably going to need to recompile it for the new platform—and the system and admin people will have to learn something new, too—but I would say that the certifications and the actual application movement would be your biggest problems.
IBM has done several things to make the mainframe a more attractive target for Linux deployments—introducing the Baby Z, standardizing IFL pricing, and other things—but do you think that still more needs to be done? Can you think of any glaring problems or omissions that IBM or other vendors need to address?
Well, when [IBM] came out with the z890, they kept their IFL processors the same price, even though speed and capacity went way up, which is good. They still—if you add an IFL to a sub-uni, the IFL still runs at the full rated speed, they’ve knocked the price of z/VM down again, by half. So if that trend continues before, too long you’re going to get a piece of hardware with z/VM included, and to me that would be a good thing, but not everybody wants to deal with that. I think it would be a good thing if people stop trying to charge the world for mainframe Linux software just because it runs on the mainframe. I think the one thing that could kill this platform is what’s come close to killing the standard platform, which is software prices gong through the roof.
Are a lot of ISVs doing this?
IBM is the only one who I know for sure has kind of maintained their pricing on the midrange software on to mainframe Linux. I would like to see more of the midrange suppliers do the same thing. I think a lot of these are priced way too high for a market where people still have to beg, borrow and steal to get an initial install up just to see if it’s going to be a problem for them.
There [are] also free trials, which some ISVs don’t let you do. Companies like LinuxCare … still let you do a free trial … Those sorts of things get people moving forward and trying things, where if they have to spend even $50 or $100, they’re not going to. I think for a long time, SuSE was enjoying free market share that they might not have had, otherwise, if they didn’t [offer a free trial period].