Some companies stay in the same business they began with, for good or ill. Others change their business model – and survive. Silicon Graphics isn’t the same it was 20 years ago, but it is alive and kicking – and today it’s focusing on Big Data.
Twenty years ago, Stephen Spielberg scared the daylights out of moviegoers with the most realistic Tyrannosaurus Rex ever seen on screen in Jurassic Park. This wasn’t Grumpy, the obvious animatronic dinosaur from Land of the Lost; this thing looked real.
Jurassic Park made a movie star out of Silicon Graphics Inc., the company whose servers were used in giant rendering farms by George Lucas’s Industrial Light and Magic (ILM), the special effects firm that did the CG dinosaurs for the movie. ILM and SGI’s rise actually began two years earlier with Terminator 2: Judgment Day in 1991 but Jurassic Park turned SGI into a celebrity in its own right.
It wasn’t just among the tech crowd. In those days, CEO Ed McCracken was hanging out at the White House with President Bill Clinton and in Hollywood with the creative community, which was unheard of for tech CEOs in the 1990s. One Wall Street analyst labeled the company ”the new Apple.”
And just as quickly, things went south. As BusinessWeek put it back in 1997, SGI executives got drunk on their own success. They jumped into new markets such as supercomputers, interactive cable TV, and digital film studios, while forgetting the basics. SGI’s new hardware was poorly done, late, incomplete, or dead on arrival. Basics slipped in marketing, inventory management, and quality control. The company lost customers and staff alike. Its former offices in Mountain View now house the Computer History Museum.
In the middle of last decade, SGI found some footing. Instead of selling $5,000 workstations, it focused on extremely high-end servers using top-end Xeons and Itanium processors. The company dumped its Unix-based IRIX OS in 2006 in favor of Red Hat Enterprise Linux, SUSE Linux Enterprise Server, and Windows Compute Cluster Server.
SGI migrated and adapted some of its old technologies from when it used MIPS processors (the company spun off the chip maker in 2002), giving SGI some pretty advanced x86-based servers. For example, SGI’s Integrated Computing Environment (ICE) built into the back plane of the computer lets chips communicate directly, instead of having to communicate via InfiniBand cables. This played a role in NASA’s Ames Research Center choosing SGI to build its Pleiades supercomputer.
But it wasn’t enough. SGI went through one bankruptcy in 2006 and then a second in 2009. The second one was fatal. Rackable Systems, a company formed almost 20 years after SGI, ended up buying out what was left of SGI for a paltry $42.5 million. Rackable then did something unusual: It adopted the SGI name, which runs contrary to most acquisitions. Usually the acquired brand disappears, the way the Sun name has largely vanished and has been replaced by Oracle branding.
Big Data Rebirth
The company is now repositioning itself as a cloud computing and big data hardware provider. SGI CEO Jorge Titinger recently blogged on the company’s vision for Big Data in 2014, which is centered around high-performance computing (HPC) and in-memory databases.
That kind of high-end computing suits SGI just fine, since it specializes in building massive, scale-out machines and supercomputers. SAP recently announced it plans to build appliances to run SAP HANA, the company’s in-memory analytics platform; those appliances will use SGI’s scalable shared memory architecture to build massive, petabyte-sized in-memory databases.
However, SGI is not entirely leaving the dinosaurs in its rear view mirror. Bill Mannel, general manager for computer servers, says that SGI’s ’90s work on graphics led to better system architectures, which suits SGI in high performance computing and big data.
“As we did work in graphics and HPC, the end result of that is lots of data. We do have customers that by the nature of their work produce lots of data. Having ways to store and analyze and visualize became part of our solution story,” Mannel said.
In 1996, SGI came up with the slogan, “Big compute, big graphics, and big data,” although the context for big data might be different. In 1996, volume, variety, and velocity were all things SGI dealt with and has for several decades now.
Rackable also has a legacy in in Big Data. It sold servers to Yahoo and Google, the two firms that started Hadoop, before Google went into business for itself making its own servers. It has also shipped hundreds of racks used in Hadoop deployments to government customers, said Mannel, doing things like in-memory computing for real-time fraud analysis.
The in-memory systems are all based on SGI’s own implementation of non-uniform memory access (NUMA). SGI’s NUMALink 6, the latest generation, first started out as a proprietary interconnect before moving to Itanium and then Xeon. This makes an SGI machine a shared memory platform running Linux out of the box on x86 processors.
SGI is offering reference implementations of Hadoop for customers completely new to Hadoop and doesn’t have that much experience, so it’s a turnkey solution out of the box. Mannel estimates 10% to 20% of customers do it this way; the rest ask for custom Hadoop installations.
“People are in a variety of different situations [with Big Data]. Some are completely in the pilot stage, not sure what they are trying to do with it,” Mannel said. “Just about every CIO right now feels pressured on what to do with Big Data. They are not always successful. Others are in full production.”
Mannel estimates about 10% of customers he’s spoken with are super users that really get the concept and are using it to help their business; 80% are still finding their way found value and are starting to invest, while the last 10% are just fiddling. Not every firm is ready for Big Data, he argues. There has to be a problem to solve. Companies should ask, “Is there something in our data to get that done rapidly and with fewer resources?” As he said, “The company has to have enough data and enough of a problem that Big Data makes sense for them. If they are around a few terabytes, there may not be enough data that’s of value. When you get up to a petabyte or more, then you are starting to see an amount of data where there could be value in following Big Data techniques.”
SGI doesn’t plan to just be a Hadoop vendor, because every x86 server vendor is doing that. It’s building on its old visualization technologies and expertise to help bring visualization to the Big Data deployments. Last September, SGI dusted off its old MineSet visual data mining technology and released a 3.0 version of the software, adding more functions for multi-dimensional visualizations of complex data sets that result from data mining analyses. The 3.0 version has its own API and runtime libraries for customization as well as access to more than 30 different file formats.
Mannel said Big Data has challenges that SGI can’t address. For starters, talent is in short supply. The McKinsey Global Institute predicts a 50% to 60% gap between the supply and demand of people with deep analytical talent within the next five years, which could mean a shortfall of up to 1.5 million managers and analysts.
In venture-capital-speak, companies need to “pivot” when their original business model fails to capture market attention. Established businesses need to change their focus too – or they disappear when the need for their specialty goes away or is subsumed in another technology. (Remember when grammar-checkers were add-on software?) If nothing else, SGI’s evolution demonstrates how one firm used its expertise to move into a new market segment.
- Gelsinger and Meyer: Two CPU Designers Who Changed the World
- The Pre-History of Software as a Service
- A Short History of Computers in the Movies: Panel Lights, Spinning Tapes, and Lab Coats
- 6 Howlingly Unrealistic Hollywood Portrayals of Law Enforcement Using Computers