x
Comms chips

Chipping Away at Cisco's ASR 9000

Details are emerging about the recently launched Aggregation Services Router (ASR) 9000 edge router from Cisco Systems Inc. (Nasdaq: CSCO), including the initial per-slot speed and the source of early confusion over the chips inside the box.

The ASR 9000, expected to ship next year, is a souped-up box that's being regarded as the long-term successor to the Cisco 7600. The 9000 is capable of running 400 Gbit/s per slot (that is, 200 Gbit/s of ingress or egress traffic at any given time), although Cisco has already conceded that early versions won't achieve that rate. (See Cisco Pumps Up the Edge.)

Now, one source familiar with the router's innards has given Light Reading a specific figure: The box will run just 180 Gbit/s per slot in its first iteration, easily outpacing the 40 Gbit/s available on routers like the CRS-1, but short of the eventual mark.

A Cisco spokeswoman wouldn't confirm that number, but did reaffirm via email that the ASR 9000 "will support an in-place upgrade path up to 400 Gbps per slot."

The nature of that upgrade remains vague. Does it involve an upgraded switch fabric? A transplantable backplane? Magic beans?

A separate source familiar with Cisco's plans says the upgrade won't require exotic technologies and that the problems with designing the 400-Gbit/s slots mostly involved power density rather than the raw speed of the signals.

No QuantumFlow of solace
On the chip front, sources from pretty much every direction say the ASR 9000's linecards will be based on EZchip Technologies Ltd. (Nasdaq: EZCH) network processors. But why, then, did an otherwise accurate early report from Oppenheimer & Co. Inc. analyst Ittai Kidron specify the QuantumFlow chip? (See Cisco's EZ 400-Gig and Cisco Plans Edge-Router Splash.)

Because Cisco said so, in a way. The ASR 9000 briefing that Cisco eventually gave to analysts referenced the QuantumFlow name. It's just that "QuantumFlow," in this case, didn't refer to the QuantumFlow Processor (QFP) that was introduced in February with the ASR 1000 router. (See Cisco Touts Chip Breakthrough.)

"Cisco is purposely causing people to be confused," one source says. "Suddenly QuantumFlow is a brand that's used for all their network processors."

Via email, the Cisco spokeswoman confirmed the wider definition: "Analysts were told that the ASR 9000 supports a member of the QuantumFlow family of forwarding engines. QuantumFlow is a Cisco silicon brand which indicates focus on high performance edge networking, with service richness and low power consumption."

For the ASR 9000, sources say the "QuantumFlow" in question is the NP-3c, a Cisco-specific variation on EZchip's NP-3 device. ("So, they can sort of, with a straight face, call it a custom processor," our source notes.) Eventual 400-Gbit/s linecards would reportedly use the next-generation NP-4.

(It's actually Marvell Technology Group Ltd. (Nasdaq: MRVL) that sells the EZchip network processors to Cisco; EZchip gets royalties. The companies announced their partnership in 2006. See EZchip, Marvell Partner.)

It's hardly new for a marketing team to take liberties with language. But the QuantumFlow incident shows that Cisco remains sensitive about admitting it's using off-the-shelf packet-processing chips.

The spokeswoman acknowledges Cisco uses other companies' chips. But neither Cisco nor EZchip will confirm the companies' relationship -- even though they're known to be working together. (See EZchip Names Cisco, Juniper.)

That's been a problem for network processor startups all along. Back around 1999, chip companies saw an opportunity to supply packet-processing devices to then-plentiful router and switch startups, but found that most customers didn't want to admit they were using externally sourced chips.

Cisco and Juniper Networks Inc. (NYSE: JNPR), in particular, had always designed their own application-specific integrated circuits (ASICs) for their linecards and switching cards. That way, all the engineering, all the know-how, and all the bragging rights stayed in-house.

Our source guesses Cisco wanted to keep up that image for the ASR 9000. "This is a pretty strategic platform, so I can see why they wouldn't want to make people think they don't control the architecture."

EZchip's sales into Juniper have been more publicly acknowledged, since Juniper represents so much of the chipmaker's revenues. Among other design wins, EZchip devices sit alongside Juniper's own packet processors in the MX series of Ethernet Services Routers, sources say.

EZchip isn't the only survivor of the old network processor days. Swedish startup Xelerated Inc. , too, has supplied chips to Huawei Technologies Co. Ltd. , while Bay Microsystems Inc. has kept afloat partly on business with the U.S. military.

Meanwhile, switch fabric maker Dune Networks claims it's got deals with systems vendors all over the place. (See Chipmakers Unfazed by Vendor Chip Shops, Xelerated Takes On Broadcom, Marvell, Bay Raises $16M, and Dune: They're Still Alive.)

Alcatel-Lucent (NYSE: ALU) remains proud of using its own ASICs (rather than network processors) in its 7750 Service Router and 7450 Ethernet Service Switch boxes. "The FP2 chipset is... the foundation for Alcatel-Lucent's roadmap to get to 100 Gbit/s and beyond, and it is in production today. No other router vendor will have samples of a comparable 100-Gbit/s chipset before mid-2009," vice president of marketing Lindsay Newell crows in an email to Light Reading. (See AlcaLu Beefs Up Its Routers.)

— Craig Matsumoto, West Coast Editor, Light Reading

<<   <   Page 2 / 2
paolo.franzoi 12/5/2012 | 3:26:13 PM
re: Chipping Away at Cisco's ASR 9000
metro,

I am replying to the guy's characterization of the launch. What the requirements for upgrade to the future capability are beyond the point. It is standard behavior for people to define what the limits of intended scalability are. That is not some special launch. That is normal behavior of just about every vendor on the market.

I actually think broad marketing of high tech enterprise and carrier products is an utter waste of time. Nobody buys anything because of a product launch. All the real outbound marketing work occurs later and individually with customers. The rest of this junk makes people happy and gets to "brand" products. It also is completely ineffectual in impacting buying behavior.

That other bit is what Cisco is very good at. Working with individual enterprises and carriers. The rest of this stuff is just all fluff for us to talk about and distract us.

Let me put it concretely. Do you think AT&T has made a single buying decision based on the product launch? Do you think this was AT&T's key decision makers first introduction to the product? (I am using AT&T as a large Tier 1 - substitute any other T1 here).

seven
Jeel 12/5/2012 | 3:26:06 PM
re: Chipping Away at Cisco's ASR 9000 Hi Brook,

Thanks for replying.
Your opinion is very welcome.

Let me make sure I understood you right.
In a nutshell, you made four separate statements:

(1) it is beneficial for a vendor to share a forward-looking roadmap, possibly way beyond the current capability and state of the product.

(2) the actual requirements for the above roadmap are not required to exist (or even be feasible). This is "beyond the point".

(3) Cisco ASR9K is not a special launch. This is a "normal" behavior of "just about every" vendor (blanket statement).

(4) broad marketing is "utter waste of time" and Tier1 carriers (AT&T example) pay little to no attention to launch information.

Not withstanding the contradiction between statements (1) and (4), it seems to me, that you deem the following model scenario to be "normal":

1. Vendor A actively markets product X based on capability Y.

2. Capability Y may not exist and is not required to exist within any contractual limits (let's say at FCS product X has 1/10th of Y).

3. Such behavior is considered "normal" and, when challenged, vendor A is allowed to point to other vendors with similar strategy - let's say vendors B and C.

Hope I got it right.
paolo.franzoi 12/5/2012 | 3:26:04 PM
re: Chipping Away at Cisco's ASR 9000
Jeel,

These are not contradictory statements. As I said, broad launches to the general public are completely useless.

These same statements as made in 1 on 1 presentation to the large carriers as these are defined as benefits to the carrier is a completely different thing. This includes such statements about the long term capabilities of the platform.

In my mind, it would be much better to remove ALL public presence of all products and provide those only privately to major customers for products like the ASR9K and only under NDA. This would help defeat any competitor strategy better than coming out to define the platform.

As to point 3, that is greatly discouraged by Tier 1 customers on an ongoing basis. The only people that allow that kind of comparison are industry analysts, the press, industry pundits and the posters here :). Using that kind of a comparison at a carrier might get an individual to be asked not to return.

seven
Beilster 12/5/2012 | 3:26:02 PM
re: Chipping Away at Cisco's ASR 9000 Yes Jeel,

Do exactly as BS says, s/he truly understands the essentials of marketing ... and be a daily participant in these enlightening LR discussion groups. Clearly this is where the sharpest marketing minds in the industry share their hone insights.
-0 12/5/2012 | 3:26:02 PM
re: Chipping Away at Cisco's ASR 9000 Well deserved sarcasm! There is no such thing as 'industry accepted practice'. Paper lunch can be accepted only from player with proven history and sizable market share. I personally inclined to take seriously (right now and in this market segment) AlcaLu and Cisco. With some skepticism - Juniper. These three are not likely to hire external PR.
If Tellabs did it I would say "Who they think they are?"
If Nortel did it - I would have a good long laugh.
If some startup did it - I would write them off and never have a second look.
<<   <   Page 2 / 2
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE