Network World Senior Editor Deni Connor caught up with Sheila Harnett, technical head of IBM Corp.'s Linux Technology Center, the day after the Linux 2.4 kernel was released. Harnett talks about the future of Linux for IBM and about its hardiness for enterprise environments.
IBM is spending $1 billion on Linux this year and saying that all its eServers will run on Linux. Why is IBM so interested in Linux?
Customers have been asking us for a couple of years how they can deploy [Linux] in existing environments. In many cases, they have existing IBM hardware that they want Linux to run on. That warrants some of our involvement in working with Linux on the Intel platform and the RS/6000.
Also, each architecture in the eServer line has different things Linux can bring to it. The mainframe-based zSeries has a unique proposition when you bring Linux to it that is different than Linux on Intel or the PowerPC.
We wanted to bring Linux onto it in the first place so environments where there are a bunch of Unix servers and zSeries machines could cooperate [on tasks]. The mainframe would run OS/390 applications; the Unix boxes would run Unix loads. Formerly, a customer would administer each environment separately.
One advantage of adding Linux to those platforms is centralization of the administration of those diverse application loads on the same box or saving time adding customers to the network. By running your Linux loads on a system alongside your native loads, you can also benefit from some of the optimizations Linux makes in having the I/O transfer more quickly inside the box as opposed to across the network.
What needs to happen to Linux to make it a solid platform for enterprise networks?
Linux can be improved in handling even larger symmetric multiprocessing configurations. The 2.4 kernel supports four to eight processors. Some of IBM's and other vendors' Unix [versions] handle up to 32 processors.
IBM is also working on high availability and clustering. Linux is very good at clustering, but needs work in the high-availability area. We are also focusing on the serviceability aspects of Linux - we are working on technologies that make Linux a better platform to debug.
Other areas are the journaled file system and logical volume management, which are two areas where Linux has been historically lacking. With the 2.4 kernel, there is the beginning of a logical volume manager that is a significant improvement. That is an area where IBM has some activity underway to be able to enhance that technology in the future.
Why is a hardy file system important for Linux?
A journaled file system is associated more with high availability of a system. When a system crashes and does a file system check on a Unix [server], it is looking for places that need to be rectified on the file system from when the system crashed. The problem is that large server deployments, in which there are gigabytes and possibly terabytes of disk space, need to wait for the system to go through every byte on every disk. A journaling file system helps keep track of file system reads and writes and logs them, so if the system does fail it knows where it left off. That's something that all enterprise operating systems have to do.
There are four open source development efforts trying to bring that capability to Linux. One is IBM's journaled file system. There is also ReiserFs, Steven Tweedy's ext3 and SGI's XFS. All four of those groups are working together to go forward to recommend a single set. To the extent that they vary, a customer can choose which [file system] they use.