The recent Oracle/NEC TPC-C benchmark, along with the 2.6 kernel, indicates without a doubt that Linux is ready for the enterprise.

During the last two years, Linux has made significant inroads in the server market, in particular in the database world. Many companies have made Linux part of their strategic IT infrastructure, with a number of organizations having fully migrated entire businesses to Linux.

However, the majority of enterprises run Linux on small, clustered systems - not large SMP boxes. With the distribution of the new Linux 2.6 kernel, coupled with SUSE/Novell SLES9 and the recent Oracle/NEC TPC-C benchmark, the market is poised to change significantly. The 2.6 kernel and the benchmark results will silence any skeptics and will reinforce that Linux is enterprise ready - on the desktop, low-end server, and large SMP box.

A Brief History

Before Linux was ready for the server market, a number of features needed to be added to meet performance and stability requirements. A significant amount of work went into adding this functionality on top of the Linux 2.4 kernel, with distribution vendors (Red Hat and SUSE/Novell Linux) adding patches to their enterprise versions of the operating system to provide a good, solid platform - based on feedback from, and joint work with, enterprise software companies like Oracle.

In the past, most Linux deployments were on 32-bit servers. Combining a fast, stable operating system with a low-cost platform was, of course, very attractive. The average server is a 4-8 CPU system with 4-16GB of RAM, a system that many people are comfortable with because it has been well tested and has performance data to support it.

Linux had a hard time running on servers larger than 16GB due to the way in which large memory management was implemented on Linux and in hardware (IA32). Because it was hard to stabilize a large memory system, scalability beyond eight CPUs on Linux was difficult; large memory and a larger number of CPUs tended to go hand-in-hand.

In many cases, the need for large-scale systems can be handled by building clusters of these cheaper, small systems. Instead of buying a big SMP system (32-way or 64-way with huge amounts of memory), a company could purchase a 4- or 8-node cluster of 4-8-way CPU systems. If an application running on top is cluster-aware and scales well, this is a great alternative to a large SMP box. Clustering technologies enable companies to leverage low-cost commodity hardware and storage to tap flexible computing resources. Additionally, with minimal upfront hardware costs, it is easy to start small and scale out as necessary.

But not everyone is satisfied with large numbers of smaller components. There is still a place for big SMP boxes. For example, some applications are built to run on large CPU systems and some data warehouse environments still require an SMP box. Until now, it wasn't possible to consider moving them to Linux. However, with Linux 2.6 coming up in enterprise distributions soon (SUSE/Novell SLES9 will be the first), it looks like the top end of the server market is within the reach of Linux.

Linux 2.6 Kernel - Made for Big Iron

Most, if not all, of the functionality that was added to Linux 2.4 by operating system vendors has been merged into the mainstream Linux 2.6 kernel. Aside from moving this code upstream, a number of improvements have been made, including support for NUMA (Non Uniform Memory Access); scheduler enhancements to handle large numbers of CPUs; more fine-grained locking; and the ability to attach more disks. The 2.6 kernel contains an impressive amount of new functionality - much of which is geared toward supporting the big iron or huge boxes. 32-way or 64-way (or even more) servers are now able to run with Linux and achieve anticipated scalability. The proof is in recent benchmark results, which illustrate exactly how much can be accomplished on a single system.

The Proof

In June 2004, Oracle announced a record-breaking TPC-C benchmark using NEC's Express5800/1320Xd 32-way Intel Itanium 2 running Novell SUSE Linux (SLES9) with Oracle Database 10g. A joint effort between NEC, Oracle, Intel Corporation, and Novell showed that Linux can scale on large SMPs. This is the best TPC-C 32-way result on Intel processors to date and the only one based on the Linux 2.6 kernel. SLES9 is based on the Linux 2.6 kernel, and every feature was used in this benchmark result. NUMA support was critical to obtaining the extra performance, as well as the improved scheduler and better disk IO throughput.

One Operating System for All

Run it on the desktop; run it on low-end or midsize servers; create clusters around smaller, cheaper building blocks; or deploy a big SMP system. Linux is one operating system that provides customers with a limitless amount of possibilities and unprecedented cost savings.