Optical/IP Networks

Cisco Unveils the HFR

Cisco Systems Inc. (Nasdaq: CSCO) unveiled its next-generation core router today, a move that some analysts believe could herald the overhaul of Cisco's entire product line, even down to enterprise boxes (see Cisco Launches HFR).

The long-awaited HFR is a multichassis "terabit" router meant to compete with boxes from Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7), Chiaro Networks Inc., Hyperchip Inc., Juniper Networks Inc. (Nasdaq: JNPR), and Procket Networks Inc.

After more than four years in development, the box is making its debut today under its real name: the Carrier Routing System, or CRS-1 (no word on whether rapper KRS-ONE has been tapped as a spokesman).

The CRS-1 truly is huge and fast, with a capacity of 640 Gbit/s in a 7-foot rack. It scales to 72 shelves rather than the 18 reported by sources, for an unreal 46 Tbit/s maximum capacity, or 1,152 OC768 ports. (Cisco reports this as 92 Tbit/s, using its usual convention of counting ingress and egress capacity separately.)

But CRS-1 wasn't intended to be just a big router, says Mike Volpi, senior VP and general manager of Cisco's Routing Technology Group. Rather, Cisco wanted to start afresh to build an IP box that would suit telecom carriers' needs for years to come. The software is engineered to produce the "permanent and continuous operation" demanded in the voice network, Volpi says. "It's designed to be Class 5-like in its carrier manageability."

Most significantly, the CRS-1 deviates from Cisco's Internetwork Operating System (IOS), the software that runs on nearly all its platforms. The new software is called IOS XR, but it's been built from scratch. The transition is analogous to Microsoft Corp. (Nasdaq: MSFT) moving from DOS-based operating systems to Windows NT, says analyst Stephen Kamman of CIBC World Markets.

Just as NT did, IOS XR could begin trickling down to lower-level systems, eventually permeating Cisco's entire portfolio, including edge and enterprise boxes. "The question is how quickly they can push that software through the product line," Kamman says.

Analyst Debra Mielke of Treillage Network Strategies Inc. notes that the amount of firepower behind CRS-1 -- including the involvement of Volpi and chief development officer Mario Mazzola -- indicates Cisco has plans going beyond this one box. "I absolutely believe that all the innovation in [the CRS-1] will go throughout the product set," she says. "They wouldn't have put all that money into [the technology] unless they were going to use it for something much more."

Kamman believes the first step will be the "Son of HFR" box, a half-sized CRS-1 intended to replace the aging GSR 12000 line (see Sources: Cisco Building 'Son of HFR'). Cisco officials won't acknowledge the half-sized CRS-1; Volpi says only that future enhancements to the platform are planned.

Critics point out that the new OS could take years to stabilize. That would put the HFR at an apparent disadvantage against, say, Juniper's T640 routers, which run that company's established JunOS operating system. "[The CSR-1 is] an interesting departure from IOS. There's the potential for [Cisco] to create more problems with their customer base," says Karen Livoli, senior product marketing manager at Juniper.

Cisco needed to make the software change someday, even if it's painful, analysts say. Because it's not modular, IOS is a step behind JunOS and other software -- something IOS XR is intended to correct (see Cisco's HFR Gets Mod).

Moreover, Cisco keeps adding to IOS piecemeal, as if it were the world's largest ball of twine. "Imagine five years from now, if they hadn't built this new software and they tried to keep IOS going. That thing would be a beast," Kamman says.

IOS XR helps Cisco catch up in areas such as hot upgrades of software and separation of control, data, and management planes. The software is based on a kernel licensed from QNX Software Systems, but tailored for the job. "We have made some pretty substantial modifications to [the QNX code] that are Cisco proprietary," Volpi says.

In terms of features, it appears IOS-XR will start incomplete, possibly reflecting the difficulty of rebuilding an entire router's worth of software. Cisco officials haven't given specifics, but it appears several features, such as IPv6 multicast and VPN support, will be missing from the initial CRS-1 release. Cisco officials say they're comfortable with the feature set, given the CRS-1's initial markets; almost nobody needs IPv6 multicast yet, and VPN software is required at the edge, not at the core, Volpi notes.

Cisco also has peppered the CRS-1 with usability features designed to make life easier for equipment operators. For example, Treillage's Mielke notes it's got flat cables, instead of the usual round ones -- a custom design that contributes to the $450,000 price tag but keeps the mass of cables manageable. No sign of a cup holder yet.

In all, it's the operational side that got the emphasis, Volpi says, noting that Cisco's engineers "did not take the macho approach of putting the most density in a half-rack chassis."

Apparently not. The CRS-1 takes a full 7-foot rack to provide 640 Gbit/s of throughput, which is on a par with the Juniper T640 (320 Gbit/s per half-rack) but lags Procket's PRO/8812 (480 Gbit/s per half-rack). As for other specifications, the CRS-1 will indeed be based on 40-Gbit/s slots -- a first for Cisco -- with an OC768c interface available. Its multistaged switch fabric is based, not on a Clos architecture, but on another well known structure called a Benes architecture.

General availability is set for July, and at least four carriers have CRS-1s running in beta tests: Deutsche Telekom AG (NYSE: DT), MCI Inc. (Nasdaq: WCOEQ, MCWEQ), NTT Communications Corp., and Sprint Corp. (NYSE: FON). Sources have frequently cited Sprint as a beta tester (see Cisco Sprints Ahead With HFR).

It appears only a chosen few carriers have gotten a look at the CRS-1. The system hasn't shown up -- other than in PowerPoint form -- to compete for any of the T640's 60 design wins so far, Juniper's Livoli says: "None of our customers have yet seen it."

Volpi says the CRS-1 project started four-and-a-half years ago. The system was developed internally, with only one acquisition -- Growth Networks -- making a contribution. "Growth Networks people built the switch fabric, although the fabric has nothing to do with the Growth Networks fabric."

— Craig Matsumoto, Senior Editor, Light Reading
Page 1 / 4   >   >>
t_jones 12/5/2012 | 1:43:54 AM
re: Cisco Unveils the HFR So the OC768 card is short range - only a 5dB link budget. Wonder what the MSRP is.
change_is_good 12/5/2012 | 1:43:53 AM
re: Cisco Unveils the HFR >>> So the OC768 card is short range - only a 5dB link budget. Wonder what the MSRP is.

who knows but i bet it will only run at 10 gig. :)
Light-bulb 12/5/2012 | 1:43:50 AM
re: Cisco Unveils the HFR Yes Yes... while I'm sure I'll take some flak for all of this I'm going to say it.
#1 IF the press release is accurate (And ALL Cisco releases are accurate...) then the actual switch capacity is 92tb/s. In a multi-stage matrix, the ingress of line card 1 can talk at full line rate to the egress of a SEPERATE line card 2 egress, while conversely the ingress from lince card 2, can talk at line rate to egress of line card 1. (This does constitute capacity since Line cards can talk to other line cards concurrently) Though a seperate question could be for hairpinning... but thats another thought...
#2 I'd be very interested in understanding the architectural limitations for distance and timing across the matrix. Do line card shelves need to be within 100m of the fabric? The Cell fabric I believe will limit the capability to push Line card shelves across a Metro area, or even a Data Center for that matter in multi-floor buildings. So for example, do we think that we could deploy a HFR in a Metro/Multi story building and maintain the Fabric and management in a central POP while having the Line Card Shelves deployed across a Metro service area, giving us a Logical router for a Metro? I believe propogation delay may cause chaos.
#3 A true Carrier OS from Cisco? I want to believe it, its time, if they have developed a substantial upgrade to the monoIOS, this is exciting news. I know I want to see this trickle down to all the routing platforms.

Ok, and a final thought. Since LR wants to negate the bandwidth usable by the platform, I suggest using a different method. (Protection 1+1 for the POS interfaces) I'm sure that if you are going to have 1+1 on the matrix, your going to want it on the line cards thus reserving bandwidth (Unless PCA) a 40gb/s slot. In which case a Carrier WILL only see 1/2 the total capacity.
Now the question on my mind...
8 port 10Gig Ethernet? Hmm, I think even LR can work that math out. So... are the slots 80gb/s? or are they 40gb/s? So the 10G ports wouldn't be able to run at line rate for all ports? Anyone?

Light-bulb 12/5/2012 | 1:43:50 AM
re: Cisco Unveils the HFR I don't think there is any immediate need for LR interface for an OC768 interface. I don't anticipate a POP that would be pushing multiple 768s across a given metro. (It may get there someday...)

Using the DWDM long-haul system is the main goal I believe in which case absolutely you want SR interfaces to keep your costs down. Though, I have a feeling that LR vs SR on a 768 POS card will be negligable when you look at the pricepoint. If you want to know my guess for the MSRP... take the OC192 x 3, should be pretty close.

Now I have to say... a OC-768 POS card is incredible.
t_jones 12/5/2012 | 1:43:49 AM
re: Cisco Unveils the HFR Good point the slots are only 40Gb per the docs. I hadn't noticed there was an 8-port 10Ge card.
Pete Baldwin 12/5/2012 | 1:43:47 AM
re: Cisco Unveils the HFR Light-bulb -- regarding the capacity, I've gone with the convention of halving Cisco's total-capacity numbers (based on Cisco's convention of "double-counting" the interfaces). The box has 1,152 maximum 40-Gbit/s interfaces, which I think most vendors would count as 46 Tbit/s.

You've got a good point, though - i'll be adding a sentence to explain the discrepancy.

As for the 8x10GE card, I'd guess it's just overprovisioned into the 40 Gbit/s slot. So, as you point out, not all 8 ports could run at 100% simultaneously. (Haven't had enough time with Cisco yet to get all the questions answered ... we're hoping to file some updates from the media event today.)
Cut_d_Crap 12/5/2012 | 1:43:47 AM
re: Cisco Unveils the HFR i would expect floundry networks (among others) to start announcing that they (1) previously announced that they actually delivered the first highest performing internet router, (2) have tons of service providers beta testing their products - although they will not tell anyone who these players are, and (3) cisco is simply following suit of their existing support for 40 gbps interfaces
durtyphiber 12/5/2012 | 1:43:46 AM
re: Cisco Unveils the HFR Just a couple of quick points....

1. I hear about "Cisco Math" all the time. Well, Cisco didn't come up with the method of "double-counting" so to speak. It was actually Cabletron that first used this method, Cisco in turn had to follow suit from a sales/marketing perspective. Just like AMD found it easier to using CPU labeling to deal with Intel on the MHZ Myth, giving them product numbers to equate to the mhz of the equivalent Intel CPU. (To head of an arguement over the MHZ Myth, I use the phrase with the definition that MHZ are not the sole arbiter of performance across all cpu architectures.)

2. An 8x10GE card would be for aggregation. Not every card needs to run LR on all ports. This is not the same as a TTM card, such as the early OC-48/OC-192 cards on the GSR. This means the card can do 40Gbs, but the aggregate port bandwidth is in excess of 40Gbs (80Gbs in this case). It is more about port density, if you need full LR, only use 4 ports. You can't fit 80Gbs in a 40Gbs slot regardless, so take your pick what you need. It's alot cheaper to run a 10GE port sub-rate than to run an OC-48 uplink. It's also more efficient and less complex than running load balancing or port channeling over multiple 1GE ports for traffic rates in the 2-5Gbs range.

Light-bulb 12/5/2012 | 1:43:44 AM
re: Cisco Unveils the HFR Craig,
I think your absolutely correct most vendors do claim (Max Interface count x Interface bandwidth). However with a Multi-stage switch, Its important to remember that it is delivering 92Tbit/s. It may really be irrelevant... as long as everyone uses the same ground rules no one get's confused. Even though I believe that system bandwidth is the summation of all data going through the matrix. Esspecially in a Core Router where interfaces are already aggregated into port interfaces.
I do believe that a industry definition should be used to help vendors market on the same playing field. Perhaps they should just say "Any port to Any port Line-rate Full-Duplex" Maybe a little asterisk somewhere on their marketting message.
I still think that any carrier will only see an effective bandwidth equal to the interfaces that are being protected. Don't know of too many carriers that think an OC768 is a commodity that can go down on any given day...

On the Ethernet side of things... I find it very interesting that Cisco would do this. If it follows the trends, it may still push Line-rate Multicasts? But I am highly skeptical because Again, a core router will be receiving traffic that hopefully has been aggregated to save backhaul costs and that traffic needs to be processed. The only benefit would be that the 8 ports would provide the CRS to do aggregation for you, maybe that is cost-effective, just need to run the numbers vs. Edge grooming/aggregation.

tsat 12/5/2012 | 1:43:44 AM
re: Cisco Unveils the HFR
Huge Fat Router?

Page 1 / 4   >   >>
Sign In