x
Routing

AlcaLu Beefs Up Its Routers

Not to be outdone by its fiercest rivals, Alcatel-Lucent (NYSE: ALU) is today announcing a major router upgrade, giving a long-awaited boost to the 7750 Service Router and 7450 Ethernet Service Switch.

A new processor gives the boxes a capacity that crosses into the "terabit router" range -- though that boast can only be achieved by counting both the ingress and egress traffic, each of which has a maximum rate of 500 Gbit/s.

AlcaLu is also adding features to the boxes, most notably deep packet inspection (DPI), firming up the vendor's heritage as a pioneer in the "service router" category.

The 7750 hasn't had a major upgrade since its introduction in 2003. And the competition has been on the move with new routers -- Cisco Systems Inc. (Nasdaq: CSCO) with its new ASR 1000, and Juniper Networks Inc. (NYSE: JNPR) with the MX family that was first introduced in 2006. (See Cisco Takes Hold of the Edge and Juniper Antes Up on Ethernet (Finally).)

So an upgrade to the 7750 and its derivative, the 7450, isn't a particular surprise -- in fact, many people have been expecting it. (See AlcaLu's Edge Upgrade.)

The interesting part is that the upgrade doesn't require a new router. It's just a matter of new line cards for the 7750 and 7450, as the switch fabric that's been in the 7750 since 2003 can absorb the extra traffic.

"We originally had told all our customers they weren't going to have to change the switch fabric to get to these speeds," says Basil Alwan, president of AlcaLu's IP business, referring as far back as Alcatel's 2003 purchase of his router startup, TiMetra. (See Alcatel & TiMetra Seal the Deal.)

Alwan adds that those customers are probably going to be surprised to see this actually happen.

AlcaLu's product launch parallels the recent launch of Cisco's ASR, in that a processor is the star of the show. (See Cisco Touts Chip Breakthrough.)

In this case it's called the FP2, and it's a chip AlcaLu designed itself, just as Cisco designed its QuantumFlow for the ASR, and Juniper designed its own chips for its MX and EX systems.

The FP2 can handle 100 Gbit/s of traffic (using the double-counting that's conventional for routers). With 10 slots in the 7750 SR-12, the largest system of the family, that adds up to what could be called 1 Tbit/s of capacity.

And in a bit of one-upmanship, the FP2 packs 112 processors compared with 40 on the Cisco QuantumFlow. (It's like neighbors competing with their barbeques and lawn mowers, isn't it?)

Cisco boasted of integrating services into the ASR, and AlcaLu isn't being left behind on that front either. Its Application Assurance card announced today adds deep packet inspection (DPI) to its routers, mirroring one of Cisco's ASR moves. An additional card for IPSec encryption is being announced for the 7750 and its smaller cousin, the 7710 Service Router.

The usual problem with these features is that, once they're activated, the router slows down substantially. Cisco's ASR 1000 demonstrates this, as the data sheets show. The QuantumFlow can normally process 20 million packets per second, but the activation of services and encryption can bring that figure to 2 million.

AlcaLu says the 7750 has been built from the start to keep routing and services running at line rate. "We didn't build a CRS, which does blazing fast core routing but little else," Alwan says, referring to Cisco's biggest router. "We continue to focus on the fact that highly classified traffic, or traffic with a lot of features turned on, should run at speed."

AlcaLu is also jumping on the "single OS" bandwagon. Juniper prides itself on having only one version of Junos in all its routers, but it's still got separate operating systems in other product lines. Cisco made a big deal of the ASR running multiple services on one OS, but competitors love to point out that Cisco's Internetwork Operating System (IOS) has dozens of versions scattered around its customer base. AlcaLu, whose IP portfolio is based entirely on the 7750, wants to point out that it's really got only one OS out there.

The enhancements to the 7750 and 7450 are due to ship in the third quarter, but AlcaLu won't tell the press what the prices are. "Pricing is upon request, but you have to have a check in hand," Alwan says.

— Craig Matsumoto, West Coast Editor, Light Reading

websterd 12/5/2012 | 3:44:12 PM
re: AlcaLu Beefs Up Its Routers We did not use a third party core. Rather, we leveraged the instruction set architecture (ISA) of Tensilica -- Cisco then did a clean sheet in-house implementation of the ISA to deliver the QFP's multi-core multi-threaded processor array. You can think of this like making a choice on the architecture for a combustion engine or a jet engine. We chose the jet engine architectural approach and then designed and built the engine from the ground up, incorporating a number of innovative capabilities in the process to make it work faster, more intelligently, and more power efficiently than any other chip in the networking industry today.
Craig, if youGÇÖd like to dig into this further, weGÇÖd be more than happy to as we believe our engineers have developed something quite innovative that will greatly benefit our customers. Just let us know. Thanks,
Doug Webster
Cisco Systems
boofritz 12/5/2012 | 3:44:35 PM
re: AlcaLu Beefs Up Its Routers Would the 7600 be the logical assumption for the ez solution?

Any ideas when we may see an update/refresh/relaunch of the 7600?

Would ez be included in new linecards in new boxes or would there also be an upgrade path for the installed base?
desiEngineer 12/5/2012 | 3:44:35 PM
re: AlcaLu Beefs Up Its Routers EZChip has said that Juniper and the world's largest network vendor are using NP-2 and NP-3 for CESR, which for Juniper means the MX. Does that mean 7600 for the WLNV?

http://www.ezchip.com/Images/p...

Juniper's website says they use their next gen Internet processor ASIC so I guess that's the IP-2.

Putting two and two together...

-desi
desiEngineer 12/5/2012 | 3:44:35 PM
re: AlcaLu Beefs Up Its Routers Craig,

You said: "I was writing under the belief that the FP2 was a single chip, not a chipset."

Does QuantumFlow have packet processing and traffic management functions in the same chip? E.g., from what I remember EZChip breaks out packet processor and traffic manager into two chips.

-desi
Pete Baldwin 12/5/2012 | 3:44:36 PM
re: AlcaLu Beefs Up Its Routers Another mea culpa -- yes, I was writing under the belief that the FP2 was a single chip, not a chipset. Which of course was wrong. Thanks for keeping me on my toes, light-headed.
Pete Baldwin 12/5/2012 | 3:44:36 PM
re: AlcaLu Beefs Up Its Routers As for EZchip -- You guys are right; I got my wires crossed between the MX (based on M-series chips plus, although I don't think juniper's officially admitted it, EZchip) and the EX.
Pete Baldwin 12/5/2012 | 3:44:36 PM
re: AlcaLu Beefs Up Its Routers You do dilute the central processor capacity. It's just that the main processor has enough cycles to keep performance (supposedly) reasonable.

Which isn't much of a "trick," i'll grant you. But to make the processor fit inside a router, especially in terms of power, probably wasn't that easy.

Didn't realize QuantumFlow used Tensilica cores. Although... hm. If you buy silicon cores from the outside and knit them into your own chip, does that still count as "home-grown?"
opticalwatcher 12/5/2012 | 3:44:37 PM
re: AlcaLu Beefs Up Its Routers Yeah, it looks like the story I heard was wrong.
The question is how close the AlcaLu/Timetra network processor is to the Treseq/Nortel one.

Of course, all startups are run by groups of people who got their experience elsewhere. Nortel certainly didn't create the idea, and it seems somewhat of a miracle that AlcaLu seems to have kept the environment that allows the team to continue their successes.
tmc1 12/5/2012 | 3:44:37 PM
re: AlcaLu Beefs Up Its Routers gocowboys,

you are only partially correct. Basil and Joe and some of the key developers/testers were rapid city. Sri was never part of rapid city, he was synoptics/bay and took over the rapid city development at bay/nortel.

Ken and Mike and the chip guys were not rapid city either, although they were not founders, they were important enough that they might as well be. They came from a NPU company called Treseq(sp?) that Nortel acquired.
tmc1 12/5/2012 | 3:44:38 PM
re: AlcaLu Beefs Up Its Routers Tera,

What you "heard" is a myth, a story, an untruth. It is not even close to accurate. The founders all worked together at Bay Networks and Nortel and many of the key guys came from the enterprise switching group that Basil was GM over. The total funding was VC and investor money of ~50M. If you look at the product design, all three boxes used the same chips, linecards, etc.

When you have super-smart and dedicated guys like Mike Noll in hw and Joe Regan designing sw features and providing the technical "glue" to make sure everything is going to work together it saves you millions. You also had Sri and Ken making sure every single penny counted and went into the product. One of the best design teams in SV and they deserve to be recognized for it along with solid leadership from Basil, Kevin, et. al.

Where most startups burn/waste money is with jr. engineers in over their heads, high priced design consultants and chip contractors, outsourced design, no one that understands the "big picture", bad financial leadership, etc.

sound like any companies from the bubble???
opticalwatcher 12/5/2012 | 3:44:38 PM
re: AlcaLu Beefs Up Its Routers "TiMetra managed to develop 7750 at a fraction of what everyone else spent on products that never worked!"

The story I've heard is that Timetra was a privatization spinoff of a spinoff of a large telecom company. If this is true, the actual amount of money they spent is probably a lot higher than the VC money spent during their last incarnation.

It is true that they didn't spend a lot of money during the telecom boom, when everyone else was spending as much money as fast as possible.
gocowboys 12/5/2012 | 3:44:38 PM
re: AlcaLu Beefs Up Its Routers tmc1,

The founders were originally part of Rapid City and acquired by Bay Networks. Rapid City was building a Switch/Router at the time.

The VC environment is very different from the inception days of TiMetra. I spent some time making the rounds recently and here are a few observations:

First, there is precious little money for telecom. There are some point investments in the valley. But many VCs are still stinging from the telecom/internet bubbles. In addition, it is excedingly difficult to sell to the carriers as a startup. So, that is a huge issue. In addition, none of the VCs will fund you based upon a plan that relies upon a single potential acquirer.

Second, funding is generally smaller. Most VCs don't want to even put in money until there are prototypes and customers. Clearly, that is not a model for developing big iron.

I am sure that Basil could get a position with a VC based upon his star power, but I doubt that a new startup would be in the offing. All things being equal, that team is doing just fine with ALU and they will probably be better served just to stay there.

The fact that ALU is beating the drum so hard about their IP products seems to indicate that everything else is probably not doing too well. But of course, Pat has a plan. :-)

tm1_queen 12/5/2012 | 3:44:38 PM
re: AlcaLu Beefs Up Its Routers "Timetra was a privatization spinoff of a spinoff of a large telecom company"
You heard it REALLY WRONG!
TiMetra was a regular (OK, not regular, but similar) VC-funded start up, just run by very smart people that kept their eyes on a ball and their egos out of the game... funded by Accel and Redpoint with the usual suspects on the Board, Wagner and Dyal. Check it out... they still take credit for it :-)
tm1_queen 12/5/2012 | 3:44:39 PM
re: AlcaLu Beefs Up Its Routers "One problem is, you'd be pumping tens of millions of $VC into a product that Cisco, juniper, Alcatel, Redback (i should have mentioned them in the story!!) and a couple others are ALREADY building. It's a tough haul, and that's why you're not seeing an explosion of switch/router startups."

Make sense, but....
TiMetra managed to develop 7750 at a fraction of what everyone else spent on products that never worked!
Pull Alwan + 20 key eng's from ALU and 7750 is dead in 2 years, OK, maybe I'm too pessimistic... give it 3 :-)
Once those 20 are out, you won't find a person in ALU that can spell "IP". "Pump" just a little of VC$ in those 20 people and watch ALU buy it again....
tm1_queen 12/5/2012 | 3:44:39 PM
re: AlcaLu Beefs Up Its Routers I guess, you have to wait and see; Q3 is not far away...
light-headed 12/5/2012 | 3:44:39 PM
re: AlcaLu Beefs Up Its Routers quote the press release: "unique new FP2 chipset"

A "set" of new chips I think not one processor like quantum flow. If you look at the picture it appears to have fewer but larger chips than previous linecard and some big heatsinks. Quite an impressive achievement as fewer chips for more performance.

Either Craig did not get briefed properly, did not pay attention or just needs to improve his reading comprehension. Typical product of USA education system maybe. ;)

No offense Craig... just keeping you on your toes.
timjjjj 12/5/2012 | 3:44:40 PM
re: AlcaLu Beefs Up Its Routers I agree double counting is unfortunate. ALU didnGÇÖt do it for 5 years. However if an increasing majority of folks donGÇÖt understand the discrepancies, being holier-than-thou becomes a negative. It sucks but sometime conforming just saves time.
yarn 12/5/2012 | 3:44:40 PM
re: AlcaLu Beefs Up Its Routers Craig,

The thing I didn't understand from the ASR launch coverage you did was that the article talked about how using a central processor (QFP) was the "trick" to scaling performance.
In hindsight this seems counter intuitive as the more cards you'd plug, the more you dilute that central processor capacity, while a distributed approach with processing on line cards performance would grow linearly. So as the 3 and 8 slot version both only have one QF processor would it be correct to assume that the best performance would be gained on the 3 slot chassis with only one line card equipped and the worst on a fully equipped 8-slot chassis? I don't know the answer, just wondering what your thoughts are on this and under what conditions you would be able to get the advertised 20 mpps.
kaka 12/5/2012 | 3:44:40 PM
re: AlcaLu Beefs Up Its Routers Where does all this chip innocations (40 cores, 112 cores) with service level processing functions leave switching chipset companies such as Broadcom/Marvell? Their chipsets are no where near to what these giants are touting?

KaKa
Pete Baldwin 12/5/2012 | 3:44:41 PM
re: AlcaLu Beefs Up Its Routers "catching up," if it's the part i think you're referring to, was meant in a marketing sense. Everyone else has put out upgrades or new families (including Redback; I should have name-checked the SE 1200 in there) ... so where's the 7750 upgrade? That's what I was talking about.

Re: Cisco, I'm pretty sure they're referencing 64 byte pakcets but would have to check.
Pete Baldwin 12/5/2012 | 3:44:42 PM
re: AlcaLu Beefs Up Its Routers Desi -- The 20 vs 2 mpps comes from Cisco's own data sheets. You get down to 2 mpps if you turn on "all baseline services combined and encryption enabled" or "numerous services and firewall enablement combined." (Quoting from the data sheet.)

I don't have a like-for-like comparison of the numbers but will try to pin something down. I agree it would be enlightening.
Pete Baldwin 12/5/2012 | 3:44:42 PM
re: AlcaLu Beefs Up Its Routers checaco:
> Well, firstly, double counting is not really conventional for all routers but for Made in Cisco.

Eh, i'd say it's become conventional. Put it this way: The headline AlcaLu sent out this a.m. has the word "Terabit" in it.

>And when calculating aggregate throughput one need not to forget about the backplane.

Yes, and thanks for calling me on that. Always true. In an effort to keep the story concise, I left out that part of the analysis.

Still, I think AlcaLu's point is that the box was ready-made for this step from the beginning. If it ain't true, we'll find out.
Pete Baldwin 12/5/2012 | 3:44:42 PM
re: AlcaLu Beefs Up Its Routers > can anyone guess, what's going to happen with the startups? are they any room for the startups to do anything posh?

If you're talking about system-vendor startups, I'm wondering the same thing. Any complex or sizable (think CRS) system is going to take more $$ than anyone's going to pump into a router startup. At least for now; maybe the pendulum swings back in 5 or 10 years, who knows.

One problem is, you'd be pumping tens of millions of $VC into a product that Cisco, juniper, Alcatel, Redback (i should have mentioned them in the story!!) and a couple others are ALREADY building. It's a tough haul, and that's why you're not seeing an explosion of switch/router startups.

That said, it may get to a point where the hardware becomes very commidified, and it's possible for a software startup to produce an interesting switch/router. Something like the Vyatta model. That's a total guess on my part and might not even be possible, but it would be interesting.
desiEngineer 12/5/2012 | 3:44:44 PM
re: AlcaLu Beefs Up Its Routers Craig,

You said that ALU caught up with JNPR. You aren't implying that JNPR has a 40G or 50G clear channel NPU. I thought their ASIC (is it an NPU) was 10G and they had multiple of them on a board to achieve a max of 40G.

Also, the number you published for cisco's QuantumFlow is 20 mpps. Is that 64 byte packets or 1500 byte packets? At 64 bytes, that's roughly 10G. At 1500, it's 240G. I'm guessing it's either i-mix or 64 byte so it would be either 10G or 40G.

That would be better characterized as catching up on capacity, but leapfrogging on capability, at least vs. JNPR, considering the 10G NPU from ALU is 5 years old?

-desi
desiEngineer 12/5/2012 | 3:44:44 PM
re: AlcaLu Beefs Up Its Routers Craig,

You mention that the performance drops from 20 mpps to 2 mpps. I searched the archives for any other info you might have published on this and I didn't find any.

Now, I'm all for this kind of real performance numbers but what did you have to turn on to make a 10x performance degradation? Not IPv6 or ACLs, I presume. DPI? IPSEC?

I presume the convoluted reporting of Basil's statement can be deciphered to mean ALU said their numbers don't degrade. Was that a like for like comparison?

Enlighten us please.

-desi
yarn 12/5/2012 | 3:44:44 PM
re: AlcaLu Beefs Up Its Routers In fact the ballyhooed Cisco QuatumFlow processor relies on commercial silicon design from Tensilica, while as mentioned on another post Juniper used EZ-chip. So ALU has no match really.

Another interesting design competition would be on router OS design. ALU managed to evolve their feature set using one OS only. That's hardly jumping a bandwagon because that feat is unmatched as well.
chechaco 12/5/2012 | 3:44:44 PM
re: AlcaLu Beefs Up Its Routers "BTW, yes that is 1T over the switch fabric. Sorry to disappoint."
Disappoint? Not at all. I just prefer straight and quite simple throughput calculation to marketing talk.
BTW. 1Tb switch fabric would not sustain 1Tb of ports at line rates. Even for unicast without any multicast. Been there, done that.
gocowboys 12/5/2012 | 3:44:45 PM
re: AlcaLu Beefs Up Its Routers I guess that this means that I can have 1Tb of traffic coming in and nothing coming out. :-)

Personally, I hate this type of performance math. A 1Tb device should be capable of full duplex at 1Tb under normal operating conditions.
timjjjj 12/5/2012 | 3:44:45 PM
re: AlcaLu Beefs Up Its Routers Only Cisco, Let's see... Juniper's MX960. True 960G. No it is 480G - Oh but that's non-redundant. Only 440G Redundant. It should be called the MX880. So you are correct Cisco double counts, just Juniper double counts plus throws a few more in for good measure. ALU was the only real player that didn't double count.

As for LR. They were more then fair. They even gave Juniper the credit for in-house developed hardware for the MX. Unless Juniper purchased EZchip and it's network processor, the statement "Juniper designed its own chips for the MX" is very gracious.

BTW, yes that is 1T over the switch fabric. Sorry to disappoint.
douaibei 12/5/2012 | 3:44:46 PM
re: AlcaLu Beefs Up Its Routers Looks all the vendor are trying to develop the latest chipset to cope with the application challenge.

Lucent and cisco has the best chipset design capability,

can anyone guess, what's going to happen with the startups? are they any room for the startups to do anything posh?

timjjjj 12/5/2012 | 3:44:46 PM
re: AlcaLu Beefs Up Its Routers Only Cisco, Let's see... Juniper MX960. True 960G. No it is 480G - Oh but that's none redundant. Only 440G Redundant. It should be called the MX880. So you are correct Cisco double counts, just Juniper double counts plus throws a few more in for good measure. ALU was the only real player that didn't double count.

BTW, yes that is 1T over the switch fabric. Sorry to disappoint.

As for LR. They were more then fair. They even gave Juniper the credit for in house developed hardware for the MX. Unless Juniper purchased EZchip and it's network processor, the statement "Juniper designed its own chips for the MX" is very graciously.
chechaco 12/5/2012 | 3:44:46 PM
re: AlcaLu Beefs Up Its Routers "The FP2 can handle 100 Gbit/s of traffic (using the double-counting that's conventional for routers). With 10 slots in the 7750 SR-12, the largest system of the family, that adds up to what could be called 1 Tbit/s of capacity."
Well, firstly, double counting is not really conventional for all routers but for Made in Cisco.
And when calculating aggregate throughput one need not to forget about the backplane. Upgrading NPU is nice and cool but wouldn't that move the bottleneck to the backplane? Or LR eats from AlcaLu palms and repeats numbers that can not be achieved in real-life networks (1T calculated without any significant number of flows crossing the backplane).
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE