451 CAOS Theory 
A blog for the enterprise open source community
Is OpenSparc all about Linux?
John Abbott, April 3, 2006 @ 6:26 pm ETI admit to being impressed by the boldness of Sun’s ‘Niagara’ UltraSparc T1 processor strategy when the chip was launched at the end of last year – eight cores, 32 threads and its emphasis on lower power consumption seemed to put it ahead of the pack. But I was surprised and a bit uncertain about Sun’s decision to turn the design specs of the chip over to the open source community as OpenSparc. What would they want with it, and what would they do with it?
Things are only a little bit clearer now that Sun has followed up on its promise, with the release of the Verilog designs for Niagara along with the programming interfaces to its HyperVisor layer microcode, the latter particularly important for those porting new operating systems to the chip. Competitor Hewlett-Packard, stung no doubt by Scott McNealy’s famous two garbage trucks quip, is putting it about that the entire thing is really about getting Linux onto Sparc – a fruitless task it points out, since 99.5% of current Linux shipments run on x86 architectures.
A quick look at the first companies to publicly show an interest in OpenSparc suggests that the embedded sector is where much of the action will be. Start-up Simply RISC, for instance, plans a cut-down version of Niagara for PDAs, desktop boxes, digital cameras and possibly routers. TTM Inc (it stands for Time To Market) provides ASIC design services. And research efforts such as RAMP and MASC are looking to synthesize the Niagara architecture on top of new-generation of field programmable gate array chips.
This is all very interesting, but doesn’t at first indicate that OpenSparc will represent a massive additional business beyond Sun’s own sales of the chip within its own servers. True, Sun created a good ecosystem based on embedded Sparc during the 1990s but that’s faded somewhat in recent years. Current Linux distributions don’t perform well in multithreaded environments and unless Red Hat and Novell begin showing more enthusiasm about Niagara than they currently seem to be, then Solaris is likely to remain the primary enterprise operating system for the new chip, tending to limit sales to Sun’s current user base.
Comments (4) Categories: Hardware,Linux
I think you’ve covered all the main factors but one: a lot of programmers really like RISC, and threading, and embedded applications; and in emerging markets, programmers do get to make some of these decisions. Remember how many of the early Linux NAS box vendors opted for RISC for its lower cost and power needs: Cobalt Networks used MIPS, for example, and RAIDtec used PowerPC. At least OpenSparc shows that Sun’s serious about engaging with the right people.
I don’t really buy it. Why does open-sourcing (and I haven’t looked at the license) the chip-set design really make a successful Linux port more likely? Unless they were getting into the business of extreme performance optimisation, no self respecting software developer is going to be checking out the transistors on the pre-fetch cache. What the developer wants is good instruction documentation and – as you point out – full documentation to microcode interfaces.
Unless…. unless, Sun was hoping to tap into a vein of open source religion, with thousand of Gnu/Linux developers stampeding forward shouting “It is an open source processor it is l33t, we … must … port … now”. Sun can’t have believed that can it? No.
I think it much more likely that Sun threw the Sparc architecture into the open, simply hoping that it would generate a little hardware/systems ecosystem out there. As you point out, it seems mainly to have resulted in stripped down Sparc implementations, rather than anything that can really feed back to strengthen the Sparc core business. Early days, of course.
FWIW, Linus merged Niagara support into the dev kernel on Sunday.
linux vs.windows webhosting.???
I wondered which should be used for dynamics sites using flash css javascript forms etc.
I was looking at this companies corporate plan and would like to have the option to set up several seperate websites for different customers
assigning each an allotment of space with seperate access by them and allowing me access is this site suitable for my needs or is it as it seems just too good to be true???