The Four Hundred

Dave Boutcher is one of IBM's top people dealing with the development and architecture of Linux on the iSeries. He has worked in the trenches turning out code; has set up environments where OS/400, Linux, and AIX operate in partitions on the iSeries; goes out and talks with customers who are interested in Linux; and is involved in planning strategies that shape the future of these seemingly incongruous operating systems.

Boutcher also plays the role of Linux evangelist, taking the gospel of Linus Torvalds to the nonbelievers, who view this approach with, let's say, more than mild skepticism. Most recently Boutcher presented his case at the Connection Conference, an annual iSeries and AS/400 education seminar. Between sessions he stepped away from his audience, but not down from his soapbox. I asked the questions and he made his points.

How long will it take before we start hearing about more companies that are finding good reasons to run Linux on the iSeries?

Linux is going to follow a similar pattern to Java. For the first year after Java came out on the AS/400, it was important to say, "We have Java on the AS/400." You didn't find a huge amount of deployment, because people were still getting their heads around it. If you look at it today, five years later, Java is a key technology on the system. Linux is following a similar growth pattern--probably a little faster than Java--but at this point in time it's very important for us to say we have that technology on the system. You can't have IBM say Linux is important and then not have it on the iSeries.

We have a lot of customers who are looking at it. We're making sales based on the fact that it is a possibility, the same as with Java when it first came out. If we hadn't been able to run Java on the system, people would have crossed us off their list. Linux is the same. We have people buying systems because they have the possibility of running Linux on it.

Where do you think Linux will find a home on iSeries? What can it do for iSeries customers?

We are seeing some Linux move into production, but we are still trying to find where the sweet spot is for Linux on iSeries. We have people using it as a firewall, but the closest thing we have to a sweet spot right now is Java programs running on Linux on iSeries.

I know of a customer that is moving significant parts of a Web-based, e-business application to Linux. That customer picked up a Java application where the ISV didn't know anything about iSeries, but it had a servlet-based application that it supported on Linux. When the customer spoke to the vendor about merging the Java application with back-end data that's on the '400, the conclusion was to run it on Linux in a partition on the '400.

Another customer had some bad experiences with WebSphere on the '400. They were gun shy, so they deployed it under Linux. These are servlet-based applications running on top of Java.

We think a lot of customers, particularly in the small to medium business space, will take advantage of the huge amount of freely available software built around Linux. One example is e-mail filtering for companies getting a lot of pornographic spam. There's a free piece of Linux software--SpamAssassin--that sits in the middle of the e-mail stream. Because it was part of the e-mail infrastructure, it needed to run somewhere that was very reliable. No one tolerates the e-mail system going down. They moved it into production on a partition in OS/400.

The value is going to be in the explosion of freely available, open-source software. You can find software to do almost anything. You can go to Google and search for Linux and mail filter or mail client or firewall. Whatever piece you search for, you're probably going to find something.

At what point are people going to start moving the core pieces of business applications to Linux?

The people who are running Java servlets on Linux are the closest to moving the core parts of business applications. But for a lot of small and medium-size businesses, it's going to be infrastructure pieces. There's a great network monitoring package called NetSaint Network Monitor, which will build a Web page that shows the status of all the systems and all the services running on those systems. When something goes down, it automatically sends an alert via pager. For a lot of small and medium businesses, this Linux package is good enough, in terms of monitoring system health, and it will save them tens of thousands of dollars.

Will companies choose this free Linux software over software designed to run on OS/400?

A lot of these decisions come down to the size of the company. The more mission-critical a piece of software is, the more they want to pay someone to be responsible for that software not working. So, for things related to running a warehouse, they will probably want to pay a vendor and get them to fix it if it stops working.

We are starting to see more "for fee" software relating to Linux. One example is Symantec. We are working with Symantec to bring a firewall to Linux for iSeries. That product is expected in August.

MAPICS, one of the traditional vendors in the OS/400 space, has a multi-tiered environment, and one tier has been designed to run in a PC. They said, "Why don't we run that in a partition in front of OS/400?" That's just one system to mange and one system to back up, instead of having an iSeries with a PC sitting next to it, which means both pieces need to up, running, connected, and configured.

For those who aren't convinced that running Linux on the iSeries is simply so that IBM can demonstrate a show of unity when it comes to its decision to back Linux, where is the real practicality?

There are two things that make me think it will be important on iSeries. Yes, you can run almost all the software in the PASE [Portable Application Solutions Environment] or port it natively. But one of the differences is that you don't have to actually do anything to make it run under Linux on iSeries. The software is out there. Either the source code is available and you just push "Go" and you get a binary, or there are Web sites where you can get Linux Power PC software that you just put on the system and it runs.

Half the customer shops I've talked to in the last couple of years are already running something on some little Linux server. The expectation for most of the other customers is that sometime within the next year somebody in their shop will try something involving Linux. Will they try it on the '400? Probably not. Some things probably suit themselves to running on a cheap, old PC. But there are some things that become important enough to migrate to the iSeries and back up, manage, and make a more managed part of the infrastructure.

Linux on the iSeries is just Linux. I mean there are no commands that run on a PC that don't run on an iSeries. There's no resistance as to where you are running it. There is going to be more and more Linux functionality moving into IT shops, just because Linux is out there and it's free.

We think a lot of the function will move up to iSeries, because when you move a function to Linux you really aren't tying yourself to any hardware vendor or operating system. If you decide to later move it--to, say, SPARC/Solaris or Intel IA-64 or whatever--the move is painless. The same source code runs on all those different hardware platforms. It may cause you to recompile, but that is a painless thing. If you just move all your user configuration files from one platform to another, they'll just work.

As a Linux on iSeries evangelist, what do you preach about the value proposition?

When IBM talks to customers about moving function to Linux, we're making a gamble that we can provide sufficient value on our hardware platforms and that the customer will run it on our stuff. Once the customer is there, he's not locked in. IBM has to compete by continuing to provide value, because it will be far easier than in the past for a customer to go somewhere else for hardware. That's not true with Microsoft Windows. It's not true with OS/400. It's not true with AIX. The cost of moving off one of those platforms is significant, and that's why a lot of people don't move.

When we talk about being able to run Linux on an iSeries, it's because we think there is sufficient value in the hardware reliability and the systems management. With our LPAR [logical partitioning] capabilities, we can do things that no other system on the planet can do, in terms of moving processing power memory into and out of a partition, including Linux partitions.

I know one example of a customer that wanted to merge three Linux servers, three [Windows] NT servers, and a SAN [storage area network] server, and they consolidated those into one iSeries system where they could run three Linux partitions, three IxS [Integrated xSeries Server] cards, and manage everything from one system.

Microsoft replacements are very interesting because there are demonstrable dollar savings.

The city government in Mexico City mandated a change from Microsoft to Linux on all desktops. It had something like 80,000 people with PCs on their desks. The savings were in the hundreds of millions of dollars.

If a company can replace 1,000 desktops, that's a big number, in terms of savings. I think you will see, over the next few years, companies replace Microsoft on the desktop. Linux distributors are very motivated to take business from Microsoft. They will make a lot of money if they can do it.

At LinuxWorld in September 2001, the iSeries was awarded the best product for integrating Linux. The reason that we won was not because we ported Linux to an iSeries. People have ported Linux to anything with a processor. What impressed the judges was the integration piece that allows multiple Linux instances to run on the system, and also move resources and processor memory between partitions. On the iSeries the data can be stored and backed up under OS/400, where it gets the benefits of RAID and mirroring. This was not just another port. It was bringing all the things that iSeries people take for granted--management capabilities such as back up and recovery, as well as integrated communications, which iSeries calls the "virtual ethernet"--to Linux.

When you talk about running Linux on iSeries it is because you can do it in a much more manageable fashion. If I'm going to move a piece of my infrastructure to Linux, doing it on an iSeries will bring a lot more capabilities--resource management, processing resources, backup and recovery, reliability, and security. If you are in a business and you just want this to sit in the back room and just run, that's why you do it on iSeries.

We want to increase the number of Linux instances you can run on the iSeries. IBM has two very good solutions for those who have a rack of PCs with lots of cables and keyboards and LAN connections. At this point the iSeries can be configured with up to 31 Linux instances running in one box. On the zSeries it really requires hundreds of Linux servers running in one box to make it efficient. We want to remove the limit of 31 Linux partitions within an iSeries box. It won't happen this year, but that's something that we have a burning desire, technically, to accomplish in the future.

What are the distinctions customers are considering for choosing one of the Linux distributors?

One is straight geography. Red Hat has the biggest market share in North America; SuSE, the biggest in Europe; and Turbolinux, the largest in Asia/Pacific. All three would like to grow into each other's space.

Another is price. Linux distributions on iSeries cost more than PC-based Linux distributions. SuSE has the most software, with 2,400 software packages on its CDs. But even if they sold one to every iSeries customer, which clearly they won't, they probably sell less copies than for Intel PCs. So the Linux distributors have to charge a little more to make up for compiling, building, and testing different packages on iSeries. Linux does tend to be price-sensitive.

Another consideration is support. When you want iSeries support, you get it under the normal contract rules. But once Linux is up and running, and you have a complaint like "My Apache Web server is kicking out this strange message every half hour," then you are not going to get support from Rochester. They do not have that expertise. So who will you call? IBM Global Services is an option, but typically you buy a support contract from the distributor.

A customer will also consider the amount of software provided. It varies quite a lot. I think Red Hat and Turbolinux are still on a single CD. SuSE is up to five CDs. It's also true that all that software is available on the Internet. You can pay less and download it yourself, or you can pay more and you get up to 2,400 software packages on your CDs.

From the standpoint of straight technology, Linux has run on Power PC processors for years. That code needed to be tweaked for running on iSeries. The existing Linux code was 32-bit, because most people were developing on Apple PowerBooks, which are 32-bit PowerPC processors.

There are 64-bits on the iSeries these days, and so IBM made a 64-bit Linux kernel. If you want to take advantage of more than 4 GB of memory, you need a 64-bit kernel. So far SuSE and Turbolinux have picked up that 64-bit kernel for distribution. Red Hat is moving slower, in terms of adopting technology on the iSeries, so its distribution is still the 32-bit version. It is working on a 64-bit kernel, but I'm not sure when it is coming out. At this point in time, its technology is not as competitive as the other two.

Most of what we have been focused on developing today has been sizing information. Customers want to know what they need to run Linux in a partition.

In our environment, we provide a lot of flexibility. That means you have to provide a lot of information. You could, for instance, assign a Linux partition a half of a processor or a tenth of a processor or whatever. So when a company says, "I expect this number of hits, and I'm running some Java servlets," you want an answer that says, "You need this percentage of a processor and this much memory." That's the number-one feedback I get: "How am I going to size this thing, if I'm going to consider it?"

The number-one source of information for iSeries Linux is www.ibm.com/iseries/linux. You can find sizing information there. It's a "cheat sheet" PDF document that you can download. We want to get to some Web-based sizing tools that will walk you through a set of questions, such as what application are you doing? How many users will there be?

One very specific customer scenario was played out using the benchmark center in Rochester to consolidate eight Linux servers into one iSeries. They were I/O-bound on their Intel servers. We set up a single iSeries system--an i840 with 808 9-GB disk drives--and then sliced it up with eight Linux partitions and a bunch of virtual disks. We fired that up and got some really outstanding numbers for I/O throughput. When Linux uses OS/400 to write its DASD, called virtual disk, it gets spread across all the disk arms, just like OS/400 normally would, which is not happening when running Linux on a PC.