Cisco Finds Nemo

Cisco Systems Inc. (Nasdaq: CSCO) is acquiring yet another tiny technology startup. This time it is paying $12.5 million in cash for Los Altos, California-based network memory specialist Nemo Systems, which is such a young company its public Website is still under construction.

Cisco's press release says Nemo has "developed leading-edge technology in the network memory space that will offer enhanced performance on Cisco's core switching platforms and service modules."

According to Cisco, once incorporated into Cisco's products, the memory technology will allow customers to scale up the use of high-peformance networking gear more efficiently. The deal is expected to close by the end of October.

The company raised its initial investment in March 2003 from founding staff, Benchmark Capital, Mohr Davidow Ventures, and several angel investors with industry experience.

So what's the history of Nemo, which will become part of Cisco's Data Center, Switching and Security Technology Group (DSSTG)? Although Web pages set up by the company are secured against unauthorized access, the company looks to be the brainchild of Stanford University associate professor Nick McKeown, a long-time fixture in the Bay Area networking startup scene who has had prior dealings with Cisco.

According to public information about McKeown, Nemo's CEO and co-founder: "He co-founded Abrizio (acquired by PMC-Sierra Inc.), which built the first commercial terabit switch fabric and was also an architect of the Cisco Systems Inc. GSR 12000 switch fabric, which uses one of his 12 patents. Nick is a Fellow of the Institute of Electrical and Electronics Engineers Inc. (IEEE) and a Fellow of the Royal Academy of Engineering."

McKeown is also a member of the technical advisory board at one of Cisco's core router rivals, Chiaro Networks Inc.. (See Chiaro Intros IP Routing Platform.) He was also an angel investor in router startup Sahasra Networks in 2001. It was acquired by Cypress Semiconductor Corp. (NYSE: CY) in 2002.

At Stanford, McKeown heads up the High Performance Networking Group, which conducts research into "core router architectures [such as] switch scheduling, buffering, packet classification and lookup, using optics in routers, and using parallelism in routers... Novel ideas in the design of the Internet architecture, e.g. load-balancing, optimal routing, deflection routing, and combining packet and circuit switching... Congestion control protocols in the Internet."

Other members of the team include: CTO and co-founder Sundar Iyer, previously a senior systems architect at a content processor firm called Switch-on; VP of marketing and business development Morgan Littlewood, who spent 13 years at Cisco and was founding president of the Multiservice Switching Forum, which is now the MultiService Forum; VP of engineering Zubair Hussain, previously of Abrizio and Chelsio Communications Inc.; and director of ASIC design Jeff Chou, previously of McData Corp. (Nasdaq: MCDTA), Cisco, and Sun Microsystems Inc. (Nasdaq: SUNW). (See Chelsio Demos 10-Gig Technology).

Cisco has been busy buying companies this year, focusing mostly on small network security companies. Its share price stood at $17.86 at the close of the market Thursday. (See Sheer Delight for Cisco, Cisco Nets NetSift for $30M, Cisco Buys Startup for $1.2M per Employee, Finisar Manages With CA.)

— Ray Le Maistre, International News Editor, Light Reading

trzwuip 12/5/2012 | 2:59:08 AM
re: Cisco Finds Nemo ntwkeng: Gourlay is not technical. He makes people believe that. And that's what good marketing is all about :) In fact, Gourlay doesn't even have a switching background. He used to work on e-commerce and then went out to the Content/Caching business unit. That BU almost shut down and he found a home with the 6000 switching BU. He has been there for less than 3 years. I am sure he is a decent guy, but he ain't all that at all. And there are hundreds of people like him in present day Cisco. He is just lucky to soak the glamor of a good BU. Right time, right place - baby!
ntwkeng 12/5/2012 | 2:59:08 AM
re: Cisco Finds Nemo No argument that Andy and Edsall are as good as it gets on the switching front. It's a shame Chambers or Mario could never get them working together. But they both have built amazing products and architectures.

Error on my part for putting Gourlay and Edsall in the same sentence; although he's the most technical marketing dude I have ever seen, but agree is definitely a marketing dude.
dljvjbsl 12/5/2012 | 2:59:12 AM
re: Cisco Finds Nemo http://klamath.stanford.edu/~s...

The presentation at the URL - Analysis of a Memeory architecture for Fast Packet Buffers may also prove instructive.
dljvjbsl 12/5/2012 | 2:59:12 AM
re: Cisco Finds Nemo http://citeseer.ist.psu.edu/66...

The paper "Techniques for Fast Shared Memory Switches" found at the above URL was written by the founders of Nemo. It should be instructive about the reasons why Cisco bought their compnay,
trzwuip 12/5/2012 | 2:59:13 AM
re: Cisco Finds Nemo "If this is rolling up to the Datacenter/Switching folks at Cisco it's the 'new' crowd of Ullal, Edsall, Gourlay, etc. They've never done a shared memory arcitecture. That was Andy Bechtolsheim and team."

Please don't mix-up Gourlay with the other folks, if you don't know the background. It's like comparing Eienstein with a lab rat. Gourlay is just an average marketing dude with probably 3 years of Catalyst 6K superficial marketing experience. Other folks are switching legends. People like Bechtolsheim and Edsall are God in the switching architecture world.
gotman 12/5/2012 | 2:59:13 AM
re: Cisco Finds Nemo Maybe a sign of a bigger better switch coming
from Cisco? I'm sure they aren't sitting there
watching Force10, Foundry and others claim Terabit switches around these conferences and roadshows? The run rate from 4K and 6k must be going back to RND.

Or could this technology be used somewhere in SAN? a lot of memory usage there, if this is ground braking stuff it could find it self a home in those boxes.
rfc1633 12/5/2012 | 2:59:13 AM
re: Cisco Finds Nemo "They'd be better off building more linecards (which is where the money is), or improving their edge and enterprise offerings (where the real money is)."

Sign, really a long way to go.
ntwkeng 12/5/2012 | 2:59:14 AM
re: Cisco Finds Nemo This company was purchased by a switching group. The switches all use dedicated buffering, generally embedded in a port ASIC. The control plane OS has nothing to do with the buffer mgmt in the hw switching path.

As far as the CRS goes, if this was purchased for the CRS it would not have been purchsed by the Data Centre group. Speaking of which- what all products does the Data Centre group have?

No argument on the advantages of Shared Memory vs Crossbar. Main issues of course are memory access speeds and depth of a shared memory architecture and sheer number of IO pins necessary on the pkg. This generally forces significant custom work resulting in rather poor yields, high chip cost.

The fastest SM switch today, I think, is ~100Mpps vs Xbar models forwarding at >400Mpps.

turing 12/5/2012 | 2:59:15 AM
re: Cisco Finds Nemo I'm confused what the router's operating system (and the OS buffer management) has to do with the forwarding plane fabric architecture. I'm pretty sure the answer is absolutely nothing.

Yes there are benefits to a shared memory fabric, and there are benefits to a crossbar fabric. But I'd be surprised if they bought Nemo for a core router fabric improvement, unless the CRS fabric is fundamentally busted. There's no need for them to be working on a CRS replacement anytime soon - they've hardly sold any CRS yet and that market is pretty dormant.

They'd be better off building more linecards (which is where the money is), or improving their edge and enterprise offerings (where the real money is).
Sisyphus 12/5/2012 | 2:59:15 AM
re: Cisco Finds Nemo I don't think Nemo has anything to do with switching technology. It's some alternative to CAMS, I seem to recall hearing.
ntwkeng 12/5/2012 | 2:59:17 AM
re: Cisco Finds Nemo I doubt it is a shared memory architecture for a switch core. More likely a high speed DRAM with very low-latency. SRAM is expensive and DRAM is gate-efficient. But the commodity DRAM vendors are not building products that suit the network industry well. (Unless you want a slow switch with interminable forwarding latencies)

If this is rolling up to the Datacenter/Switching folks at Cisco it's the 'new' crowd of Ullal, Edsall, Gourlay, etc. They've never done a shared memory arcitecture. That was Andy Bechtolsheim and team.

rfc1633 12/5/2012 | 2:59:17 AM
re: Cisco Finds Nemo Well, I will definitely agree with you that "They've never done a shared memory arcitecture". This is a truth, a result, but why? Did you ever think about CiscoIOS's architecture and how maladroit of its buffer mgmt?

The one who take share memory back, is IOX, a QNX core-based system. Multi-chassis CRS-1, will have a leveled crossbar fabric. Imagine how many queues needed when a packet go across the box!!!

Using crossbar, we must cook the header first and then move the whole packet from ingress mudule to another module, accross the fabric. Then maybe the hdr need to be cooked again on egress module, and get another specific function/asic invovled. Then introduced one more MemIO. If you take a indepth look into GSR LC's stucture, and count the numbers of buffers invovled during a packet go through GSR, you will worry about delay for sure.

For share memory, during the whole packet across path, we need only one time memory read/write. The only thing need be moved here and there is Hdr. Wihle hdr rewriten done, just take the data part patched after the final cooked Hdr. You will definitely get shortest delay.

The more features a box/module perform, the more share memory arch. needed.
rfc1633 12/5/2012 | 2:59:17 AM
re: Cisco Finds Nemo Not very sure about Nemo's detail. But Cisco should willing to take place crossbar switch fabric by Nemo. Looks share-memory switch fabric come back again.

Using Crossbar, there must be a lot of Queues before and after Crossbar, this will increase delay dramatically, which is really bad thing to a High Performance Core Router. Think about Procket's architecture, you will get this point.

voyce_overipee 12/5/2012 | 2:59:20 AM
re: Cisco Finds Nemo But it's a self-feeding loop now isn't it? Lightreading used his name because you've heard it more, ergo it gets heard even more.
goundan 12/5/2012 | 2:59:20 AM
re: Cisco Finds Nemo What kind of networking memory chips did Nemo pioneer? CAM, SRAM, some kind of low latency DRAM? Can anyone give an idea why Cisco would pay big bucks to acquire them?
sjd6 12/5/2012 | 2:59:21 AM
re: Cisco Finds Nemo is now an insignificant pimple on cisco's a$$
Scott Raynovich 12/5/2012 | 2:59:21 AM
re: Cisco Finds Nemo Google war:

Nick McKeown + networking = 101,000 hits

Sundar Iyer + networking = 862 hits
Balet 12/5/2012 | 2:59:22 AM
re: Cisco Finds Nemo ...Lightreading?
Scott Raynovich 12/5/2012 | 2:59:23 AM
re: Cisco Finds Nemo Nick,

Thanks for the clarification. WE used your name in the summary line cause you're probably the more well known of the two of you. It's clear that you were cofounders.

digits 12/5/2012 | 2:59:23 AM
re: Cisco Finds Nemo I know it was "The Incredibles"...

But for Cisco -- who else does it have its eye on?
nickmckeown 12/5/2012 | 2:59:24 AM
re: Cisco Finds Nemo Nemo Systems was co-founded by Sundar Iyer and me. Your article gives me more credit than I deserve: Sundar was the original brains behind the network memory technology.
Sign In