Acquisitions of SGI, Sun Heading in Opposite Directions for HPC
as published in HPCWire
Sun and SGI have followed similar trajectories throughout much of their linked histories — parallel parabolas that blazed through several common markets,
including technical workstations and HPC, from their explosive Silicon Valley launches in the early 1980s to their financial zeniths in the 1990s.
Ultimately each company failed to achieve a sustainable orbit, and the last ten years have seen their slow, fiery reentries and projected crashes
to terra firma.
This duality of existence extended into each company’s apparent ultimate splashdown, the now-closed acquisition of SGI by Rackable and the announced
but still pending acquisition of Sun by Oracle. But this is where the twin contrails could finally diverge, at least for HPC.
On the one hand we have Rackable — or should I say SGI? — for Rackable has officially adopted the SGI brand. For HPC customers of the new SGI, the
news looks good. Although there may be some consolidation where x86 lines overlap, the new SGI is going forward with the core HPC and visualization
business, including most of the acquired employees. SGI will continue to fulfill product and service contracts, including the $40 million DOD deal
from February. The Rackable-SGI deal is reminiscent of Tera’s acquisition of Cray assets from the selfsame SGI in 2000.
In that deal Tera took on the Cray brand and began the laborious process of bolstering the company’s brand, technology lines, and revenues in HPC.
The acquisition essentially saved Cray in HPC, and Rackable may have just done the same for SGI.
Although the deal is yet to close, Oracle-Sun seems to have the opposite prospects. Oracle is a software company, and its management has made no secret
of the fact that this acquisition is about software, notably Solaris and Java. An outside observer might notice that Sun has been known to dabble
in hardware as well, including servers, processors, networking and storage for HPC.
Will Oracle keep any of the hardware pieces? They would have to be very selective if so. This is unlike Cisco’s move into servers for virtualization,
in that Cisco at least already had familiarity with building things. The operational challenge of transitioning from software to hardware is not
to be taken lightly. There are inventory issues, parts shortages, and spares for the field. In HPC there is burn-in, benchmarking, and acceptance
testing. I just don’t see it in Oracle’s future.
Hardware notwithstanding, there is also a software technology owned by Sun that is significant to HPC but hitherto unmentioned in the acquisition discussions.
What is to become of Lustre? With pNFS approaching GA, this is a critical year for parallel file system technologies, and Lustre’s roadmap is now
This is where Sun’s and SGI’s paths ultimately diverge. While SGI has a long road to recovery ahead of it in HPC, at least there is a road. All Sun
seems to have is an off-ramp.