The word coming out of the Sun portion of software giant and seemingly enthusiastic hardware supplier Oracle is that the axe has fallen on the company's HPC group.
While details are scarce — just the way Oracle likes it — El Reg hears from sources familiar with the matter that layoffs took place last week, with most of the HPC sales force let go and the remainder redeployed peddling Oracle's Exadata V2 data warehousing and online transaction–processing clusters.
A member of one software company in the HPC space, who requested anonymity because of partnerships with Oracle, contacted El Reg to say that résumés of Sun employees who had been working on HPC projects were flying into his email box. It's unclear how many people were ever in Sun's HPC organization, which spanned systems, storage, switches, and software, and it's equally unclear how many were left after Sun's and Oracle's several rounds of layoffs in the past few years.
As usual, Oracle did not make an executive in its Server and Storage Systems group available to comment on the layoffs nor any restructuring Oracle might have done to repurpose former Sunners inside Oracle. An Oracle spokeperson refused to comment on the matter.
Even before Jonathan Schwartz took over as president at Sun in 2004, the company had been jonesing to get back into the HPC game, which used to generate about $2bn in a year in sales for Sun, including workstations, servers, storage, networking, services — and the broadest definition of high performance computing you might allow.
In 2006, Sun was even getting some respect from the US Defense Advanced Research Projects Agency, which sponsored a bunch of forward-thinking, petaflops-scale supercomputer projects and awarded cash to Sun, IBM, and Cray; only Cray and IBM made the final cut in November 2006, but Sun soldiered on with the launch of the "Constellation" clusters for media streaming and HPC announced in June 2007, complete with Sun's own monster InfiniBand switches. Sun finally seemed to be getting serious about supercomputing. Perhaps just a bit too late.
This is a sad ending to an HPC story that started out with well-regarded workstations and then servers running Solaris, and evolved into a Sun that bought the Cray 6400 business and brilliantly turned it into the "Starfire" Enterprise 10000 server line that fueled the dot-com boom and made Sun a darling of Wall Street. Sun also ate carcasses of Thinking Machines and Kendall Square, and had plenty of people to work the HPC field.
If only Sun's techies had delivered on the promises they made in 1997: the UltraSparc III line of servers — over 1,000 processors in a single system image — and the "WildFire" high-speed system interconnect to link multiple machines together to boost scalability further and create a shared memory system more tightly coupled than the global shared memory systems sold by Silicon Graphics.
Had the "Serengeti" UltraSparc III machine panned out and WildFire interconnects linked multiple machines together, Sun might have been the one eating Oracle and not the other way around. But once all those supercomputing people joined Sun in the late 1990s and early 2000s — perhaps a coincidence, perhaps not — the company began a nasty habit of promising a lot more than it could deliver in chip and system designs. Now that Sun is part of Oracle, Sunners don't make any promises.
But Oracle is still interested in making and selling servers. And next week, on August 10, John Fowler, the top ex-Sunner at Oracle and the executive vice president in charge of Oracle's Server and Storage Systems group, is going to host a webcast explaining Sun's systems strategy.
Sun is just about due to deliver its 16-core "Rainbow Falls" Sparc T3 chip for Sparc T series servers with one, two, or four sockets. That chip was expected to run at 1.67GHz, with 16 cores on a single die and eight threads per core.
In June, Oracle already revamped its significantly streamlined Sun Fire x64 server lineup with the latest Xeon 5600 and Xeon 7500 processors from Intel. But who knows, maybe Oracle's killing off of the Opteron-based servers, which Oracle admitted to customers in May but has refused to confirm, was nothing more than a negotiating tactic for better pricing and technical help from AMD.
It looks like Oracle yanked HP's Solaris OEM contract to get the server maker's attention and to renegotiate the terms of a new deal more to Oracle's liking, and it would not be at all surprising that the company is doing the same thing with AMD and we'll see some Opteron-based machines coming down the pike. (It is equally likely that Oracle really meant it and there will be no Opteron servers. You need a bloody oracle to divine what Oracle will do next.)
And as El Reg previously reported in early June, Fujitsu president Masami Yamamoto said that the two companies are working on a new contract to work together on the midrange and high-end Sparc Enterprise M systems, which for the past several years have been rebadged as Sparc64 machines designed and built by Fujitsu. Sun has not built its own midrange and high-end servers since the dual-core UltraSparc-IV+ chip was in the field, and was hoping that the Fujitsu machines would be a stopgap between the "Millennium" UltraSparc-V, which was killed off in April 2004, and the "Rock" UltraSparc-RK, which was supposed to be in the field in late 2008 but which was put out of its misery around June 2009 while Oracle was in the process of trying to buy Sun. Neither chip saw the light of day in a system.
Fowler could enlighten customers about what the Fujitsu partnership will look like in the future, and what products the two companies are working on. Then again, the Fujitsu deal may still be in the works and Fowler may not be able to say anything about it. ®