But Wait, There's More

Another Possible Fate for Sun: General Electric

Last week, when I was ranting about the possible fate Sun Microsystems might be facing if it encounters insurmountable difficulties in transforming itself into a supplier of components of grid and utility computing, I said that if Sun stumbled hard it would likely disappear into the gaping maw of partner and Japanese industrial giant Fujitsu with hardly an audible burp. There is, of course, a more interesting possibility.

When Sun turned on the Sun Grid utility earlier this year and began discussing in great detail its vision and future products for utility computing, I said that Sun aspired to be the General Electric of the grid computing era. As it turns out, if Sun is right about utility computing and if it runs into economic difficulties, GE itself would be something of an ideal owner of Sun.

Think about it. General Electric was founded in 1876 by Thomas Edison to provide the infrastructure for the electricity grid necessary for commercial and consumer lighting and the use of electricity for appliances as well (which GE would eventually make, starting with the electric fan in 1890). GE is also a key player in other infrastructure markets--it makes railroad engines and aircraft engines. GE was in the computer business decades ago by virtue of its ownership of a piece of RCA Computer, and it also had a massive embedded controller business. And, its General Electric Information Service (GEIS) has been a backbone for corporate computing, and was so before the Internet became commercialized a decade ago as an alternative. GE could decide to be a supplier of components for the utility grid--if it thinks there is money to be made in either running a grid or being a key supplier. With a current market capitalization of $386 billion and sales of $152.9 billion in 2004, GE could easily absorb Sun. In fact, with about $13 billion in cash, GE could buy Sun without diluting its stock.

Solaris-Sparc Bests Lintel Iron on OLTP Tests

Sun Microsystems is allergic to the TPC-C online transaction processing (OLTP) benchmark test, but it nonetheless wants to prove its Solaris boxes are as good at OLTP jobs as other machines. So Sun did what a lot of server vendors have done over the ages: it came up with its own OLTP benchmark and it ran the test on one of its own machines and on a competitor's box, and then proclaimed itself the winner.

Specifically, Sun created an OLTP test it is calling iGEN_OLTP_v1.5 and ran it on a Sun Fire V440 server with four 1.6 GHz UltraSparc-IIIi processors; the server ran Solaris 10. Sun also ran a test on a four-way Dell 6650 server with 3 GHz Xeon processors (very likely 32-bit chips, but Sun doesn't say); this machine ran Red Hat Enterprise Linux 4. Sun says the Sun Fire box had a peak performance of 103,706 transactions per minute (TPM) gauging only the medium transactions running on the system. (There were also heavy transactions, which impacted the performance of the system running those medium transactions.) This compared favorably to a peak performance of 94,347 TPM on the Red Hat-Dell box. Somewhat inexplicably, Sun said that with 2,100 connections to the server (presumably this means end users), the Sun box ran at peak performance, but the Dell box became saturated and performance dropped to 68,005 TPM. Both servers were apparently running the open source MySQL database.

Here's the description of the OLTP benchmark Sun used: "iGEN_OLTP_v1.5 is a benchmark for measuring the throughput and response times of an OLTP database workload. This workload was based on customer applications and has a mix of light-weight and heavy-weight transactions. This benchmark is also used to show the performance of competitive systems under heavy load. The benchmark has two performance metrics: Throughput in tpmM (Medium transactions per minute) and average medium response time in seconds." I am trying to get a more detailed description of this benchmark and the source code to see what really is.

I think a new OLTP benchmark is a good idea--and so does the TPC, incidentally, as I have written about several months ago. (See "New TPC Benchmarks Are on the Horizon" from the October 7, 2004, issue of this newsletter.) However, throwing out a new benchmark without the specs, without an independent audit, and without giving the competition--in this case, Red Hat and Dell--a chance to make sure the tests were kosher is not particularly friendly or fair. That said, the animosity IBM created when it ran benchmarks on its AS/400 minicomputers and competing equipment from DEC and Hewlett-Packard in the late 1980s is what eventually lead to the creation of the TPC suite of benchmarks for OLTP workloads. Some good beyond PR could come from Sun's cheeky move.

KT Picks Solaris Iron for Telecom Infrastructure

Korea Telecom, which is now calling itself KT, this week inked an infrastructure and services partnership with Sun Microsystems to help it build a telecom operation support system based on its Sun Fire servers, Solaris, the Java Enterprise System middleware stack, and the Java programming language. Sun and KT will be working to create a system to manage the converging wireless and broadband networks that KT is selling like hotcakes in Korea. This system will have a Web-based front end, and it will run on a collection of Sun Fire Enterprise 4900 and 6900 midrange and Enterprise 25000 high-end Sparc servers as well as on Sun's Opteron-based V40z servers. Financial details of the deal were not disclosed.

SCO Partners with White Box Server Makers to Create Preconfigured Unix Servers for Resellers

Back in the dawn of time, before Linux and Windows were part of the commercialized server market, software companies that peddled turnkey applications to small- and mid-sized businesses more times than not created their applications on two platforms: SCO Unix or Novell NetWare. Back then, companies that bought solutions usually didn't even know they were running Unix or NetWare on their X86 servers, and quite frankly, none of them should have cared so long as the solution worked and the platform was stable.

A lot has changed in the past decade and a half, but The SCO Group, which is still trying to sell Unix for X86 servers, wants to get back to where companies bought servers with Unix already installed. It is going to have to take a different route to get there. This time around, SCO is partnering with so-called white box server makers--which means they are generally smaller regional or national server suppliers without the big brand names--so they will preconfigure SCO's UnixWare and OpenServer Unix variants on servers on behalf of the SCO resellers that want to peddle turnkey solutions. SCO has chosen Terian Solutions, a white box maker based in Houston that makes Xeon and Itanium servers, and DTR Business Systems, another vendor based in Walnut, California, that sells X86 servers. Both vendors, incidentally, sell Linux boxes, and DTR also sells Windows solutions.

Cray Posts Big Loss on Weak First Quarter

Supercomputer maker Cray has posted a loss of $21 million in its first quarter ended March 31 on sales of $37.6 million, down 11 percent. Cray posted a net loss of $3.8 million this time last year, and Wall Street is pretty jumpy that the company had such a big loss, which appears to be driven by weak sales and unexpectedly high research and development costs. In the past year, Cray rolled out three new product lines, and had hoped to be cashing in on them profitably by now. Jim Rottsolk, Cray's chairman and CEO, said the quarter was also impacted by manufacturing issues and it had fired IBM as the foundry for certain components. Delays in component shipments for its supercomputers have held down sales in past quarters; firing IBM cost Cray $1 million, and writing off other component inventories and "manufacturing variances" accounted for a total of $5 million in charges in the quarter. Rottsolk said its product inventory rose to $103.2 million in the quarter as the ramp of the X1E, XT3, and XD1 lines continued, and $37.3 million of that inventory was installed at customer sites awaiting customer acceptance so they can pay the bill and Cray can recognize the revenue. He said further that Cray has a $127 million backlog of orders for the next 12 months. Cray's difficulties in making money on its substantial engineering efforts and acquisitions (the XD1 was designed by the former OctigaBay) have made investors unhappy, and Cray's stock is now trading in the range of $1.50 a share, down from $8 a share a year ago. With a market capitalization of only $131 million, Cray is a prime target for a hostile takeover. The company's engineering knowledge alone is probably worth several multiples of the company's current value. Ironically, Cray paid $15 million in cash and $100 million in stock to acquire OctigaBay a little more than a year ago to get its hans on that company's innovative Linux-Opteron supercomputers before they hit the market.

In a separate announcement, Cray said it has hired Margaret Williams from rival IBM to become its senior vice president of engineering. Williams is a 20-year veteran of the supercomputing market, and she used to manage IBM's AIX and parallel computing software development. She was most recently a vice president of database technology within IBM's Software Group, but was also in charge of IBM's team at the Maui High Performance Computing Center in Hawaii, which is a supercomputer center for the U.S. Air Force that has two AIX-Power parallel SMP clusters as well as a Linux-X86 cluster. Williams will take over Cray's entire engineering effort, including the X1E, XT3, and XD1 lines as well as the advanced research projects code-named "Cascade" and "Rainier." Cascade is a long-term project that Cray has underway to reach the sustained petaflops (1,000 teraflops) performance range by 2010. Rainier is a project that is supposed to emerge in the 2006 to 2007 timeframe, and is based on heterogeneous computing, which probably means a mix of Cray servers (which have different architectures) working in concert on jobs.

OLAP Report Says OLAP Market Grew 15.7 Percent in 2004

According to the analysts at British market researcher The OLAP Report, last year was the best one since 2000 for the online analytical processing (OLAP) subset of the data warehousing software market. OLAP Report reckons that companies engaged in this market segment had $4.3 billion in sales last year, up 15.7 percent. This was ahead of predictions, and well ahead of the single-digit growth the OLAP market had from 2001 through 2003. OLAP processing got off to an explosive boom in 1996, with $1 billion in sales, and hit above $2.5 billion in 1999. The company is projecting good growth, with sales approaching $6 billion by 2007. That growth will come even after several waves of consolidation and a lot of pricing pressure from Microsoft, which started embedding OLAP services in SQL Server in 1999, tweaked it in 2000, and has been using price to compete as other vendors add functionality.

The consolidation in the OLAP software industry is profound, with the top five vendors--Microsoft, Hyperion Solutions, Cognos, Business Objects, and MicroStrategy accounting for 76.5 percent of worldwide sales in 2004; the top 10 vendors account for about 95 percent. OLAP Report reckons that about a third of Microsoft's SQL Server sales should really be attributed to OLAP and extract-transform-load (ETL) tools. Hyperion's products, which have been ported to the iSeries, are holding their own against Microsoft so far, but the company has seen a steady drain on revenue and market share since Microsoft entered the market.

IBM Buys Open Source Middleware Supplier Gluecode Software

Having seen the writing on the wall now that the open source JBoss Web application server is J2EE compliant and growing rapidly in popularity and its WebSphere software is overkill for many small and mid-sized businesses, IBM made an interesting move last week when it acquired privately held Gluecode Software, a provider of support for the Apache Geronimo Java application server based in El Segundo, California.

Within the next few weeks, Gluecode was getting ready to launch an enterprise-class version of its Joe application server, which is based on the Apache Geronimo server, which is itself an open source alternative to JBoss, WebSphere, WebLogic, and other J2EE-compliant Web application servers. Like other companies trying to make money off of open source projects, Gluecode offers packaging, support, and other services for Apache Geronimo. IBM needed a less sophisticated and less costly Web application server, and Gluecode is peddling one that offers support that costs from $500 to $2,500 a year. How IBM will integrate Gluecode with the iSeries remains to be seen, but in keeping with its eServer philosophy, it is hard to imagine the Joe app server won't eventually be an option on the iSeries. IBM has been using the Apache Web server at the heart of WebSphere for years, and recently contributed its Cloudscape database (written in Java) to the Apache Foundation's Derby Project. IBM plans to contribute to the Geronimo project and to make Gluecode's software available for free. Gluecode will be integrated into IBM's Software Group.