“We need more programmable networks and high level APIs for them—we
should have done this years ago and companies such as Cisco and Juniper
will all do this,” Bechtolsheim told a gathering of engineers outside
the ballroom. “But that’s not OpenFlow, which is a protocol for a much
lower level of the network,” he said.
OpenFlow emerged from an
initiative at Stanford Unversity to define a ground-up way to simplify
large networks by running more of the work as applications on servers.
Google described its work creating an internal OpenFlow network at a
“Google’s talk was at an OpenFlow event but it was
not about OpenFlow,” said Bechtolsheim,. “now founder and chairman of
switch maker Arista Networks. “We support [OpenFlow] because we had a
customer who wanted to try it, but no one is really using it,” he said.
also took a swipe at competitors such as Cisco Systems who develop
their own ASICs and custom chips. Arista uses only merchant silicon in
“Cisco still believes they can win the war by
designing their own proprietary chips and they have the volume to do
it,” said Bechtolsheim after his talk. “But it still sounds like the old
Sparc arguments to me,” he said, referring to CPUs developed by Sun
Microsystems which he co-founded.
ASICs cannot achieve the
density and clock rates of full custom designs, he said. He predicted
merchant 28-nm switch chips will be available by 2015 that support up to
256 10G ports and 100G line rates.
“Merchant silicon market will
gain market share,” he said. “The next two iterations [of switch chips]
will see very rapid improvements in cost and performance."
What's Andy smoking these days? Apple would counter his merchant silicon argument and who's developing full custom chips these days except Intel and AMD. Most merchant silicon in networking are developed using standard cell ASIC libraries.
He's got his head in the clouds these days. Arista will get absorbed into some other networking company if they keep reselling Intel reference designs. Using merchant silicon is good if you want to get to market quickly but you can't hold your lead relying on suppliers. All your competitors have access to the same silicon.
It might be more accurate to say the *market* gave up on SPARC.
SPARC is a RISC processor, and an outgrowth of rethinking the architecture of a CPU. Previous generations had all been CISC architectures, with huge instruction sets. But realization set in that most of those instructions weren't actually used because compilers didn't generate code that used them. Why not design a CPU with only the basic instructions from which others could be constructed, and concentrate on making them as fast as possible, then let compilers do the work?
Sun created the SPARC. HP had Precision Architecture. DEC created the Alpha. AMD had the 29000. But takeup was limited to server space, and steady improvements by Intel ate into any advantages RISC possessed. CISC was just as fast as RISC, and was *cheaper*.
Sun wasn't the only one who backed off. DEC no longer exists. HP moved away from PA to systems based on Intel architecture. AMD concentrated on being an alternate supplier of Intel architecture chips.
Backing away from SPARC didn't kill Sun. Being unable to compete did. You can make an argument that developing SPARC used resources Sun might have better applied elsewhere. If I'm a customer looking at Sun servers, and I see SPARC and Intel-architecture Opteron models, they can do the same things, and Opteron is cheaper, which way do I jump?
SPARC may well be technically superior, but customers aren't buying technical superiority - they're buying bang for the buck. If you offer a product that costs more, you better have something the *customer* will see as justification to pay the higher price. SPARC wasn't it.
I'm not disagreeing with your analysis, but the reason DEC no longer exists is that Intel bought it as part of the settlement of the patent lawsuit brought by DEC against Intel.
Intel had no reason to keep producing the Alpha or keep the DEC name past the agreed period. It pays to be the 880 lb. gorilla.
Well, the Alpha no longer exists because Intel bought it and stopped making it.
DEC's demise is a little more complicated, as DEC had been selling of parts trying to stay alive before what was left was acquired by Compaq.
Far that matter, DEC competitor Data General suffered a similar fate. They made the AViiON line, originally based on the Motorola 88000 RISC processor, and shifted to Intel when Motorola dropped the 88000.
DG was eventually purchased by storage vendor EMC, who bought them to get the CLARiiON storage system, and promptly stopped making the computer.
But yeah, it pays to be the 800lb gorilla. It's yet another acquisition made to buy and kill off a potential competitor.
It appears that Andy can get headline articles written based just on his hallway uterings!
His comments on Sparc were referring to the vertical integration of silicon, system h/w and software. This is difficult for any company to keep delivering on. Being able to use the best of current breed of silicon and differentiating on the software is where the system companies are finding success.
Apparently on the marketing front Arista supports the buzzwords of SDN and actually supports some version of OpenFlow.
But privately Andy believes OpenFlow is too limited, low level and incompatible with today's networks to gain traction.
He does believe there will be more mangeable, configurable nets using some form of SDN with each major company probably doing their own API thing for the next several years.
Let's see Apple has it's own HW (silicon, boards, and I/O specs) and SW (OS, and applications). Sounds vertically integrated to me. Only thing missing is manufacturing but if foxcon has more problems, this many change.
Seems the pendulum has swung back the other way and the industry is still stuck in 'core competency' mode -- or rather chucking competency and innovation.
Just by way of reminder, silicon photonics is very much an SOI-based technology. See http://www.advancedsubstratenews.com/tag/photonics/ for articles explaining the role of SOI in photonics by Intel, IBM, Luxtera, Sony and more.
Join our online Radio Show on Friday 11th July starting at 2:00pm Eastern, when EETimes editor of all things fun and interesting, Max Maxfield, and embedded systems expert, Jack Ganssle, will debate as to just what is, and is not, and embedded system.