x
Optical/IP

Cisco Scores CRS-1 Customers

Cisco Systems Inc. (Nasdaq: CSCO) is all over the news map this morning, as it has four big routing announcements involving its flagship core router, the CRS-1. The company has revealed some CRS-1 customers, a partnership with Fujitsu, a strategic answer to Juniper Networks Inc.'s (Nasdaq: JNPR) Infranet Initiative, and its long-awaited half-sized version of the CRS-1. Here are the details on each development:
  • Cisco is finally naming some users of the CRS-1, after much speculation by analysts (see Cisco Stumps for CRS-1).

    The marquee name is Softbank, which is "going through a massive deployment of CRS-1s" for its YahooBB service, says Suraj Shetty, director of marketing for Cisco's routing technology group. Cisco won't give specifics, but Shetty says the number of chassis being used here is in the "strong double digits." The deployment has already begun and will be done in "months, not years," he says.

    That's the only solid telecom deployment Cisco is citing. The Pittsburgh Supercomputing Center (PSC) is another customer being announced today, noteworthy because it is using the OC768 interfaces -- which, incidentally, performed quite well in a recent test sponsored by Light Reading (see 40-Gig Router Test Results).

    Other CRS-1 customers being announced are Japan's National Institute of Informatics and Telecom Italia SpA (NYSE: TI), the latter being just a trial customer.

  • Cisco has also set up a partnership with Fujitsu Ltd. (OTC: FJTSY; Tokyo: 6702) for high-end routing in Japan. The CRS-1 essentially becomes part of the Fujitsu roadmap, as the company will be pitching it as a next-generation cog for high-end customers. While similar to core router reseller agreements, such as the Nortel arrangement with Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7), the Fujitsu-Cisco partnership appears to be unique because Fujitsu is a router vendor itself and has been pitching a different company's routers for the core in Japan.

  • Cisco is also launching a program that sounds more than a little similar to Juniper's Infranet Initiative. But Cisco officials insist they're hunting bigger game than Juniper is.

    Cisco's "IP NGN" project -- letters standing for Internet Protocol (IP) Next-Generation Network -- is getting formally launched today. The idea is to create a network that responds to what the user is doing, in hopes of creating more loyalty among service provider customers.

    IP NGN is part of Cisco's ongoing effort to be taken more seriously in telecom circles. Most aspects of the company's CRS-1 core router, including the IOS XR modular operating system, are intended to curry favor with service providers. "Almost 50 percent of our R&D is going toward service provider development," says Cisco's Shetty.

    Identity is a key piece of IP NGN. The network would be able to recognize a user as he shifted from cell phone to PDA to office-bound LAN connection, and would likewise alter quality of service (QOS) and other parameters depending on the application being run. Voice and video traffic would receive priority over delay-insensitive data traffic, for example. "The network would provide resources to the users on demand," like a "customized tollway," Shetty says.

    Partly, the goal is to engender user loyalty. But the IP NGN would also let carriers offer premium services to customers based on their usage patterns, creating new possibilities for generating revenues.

    Cisco tipped its hand in this area with the acquisition of P-Cube, which developed deep packet inspection systems for distinguishing traffic content on the fly. In fact, Cisco is using P-Cube's Service Control Engine line of appliances as examples of what the IP NGN is trying to do (see Cisco Plucks P-Cube for $200M and Cisco Reroutes Traffic Management).

    It all sounds an awful lot like what Juniper is describing with its Infranet. But Cisco insists the Infranet is a subset of what the IP NGN would accomplish.

    "The way we understand it, the Infranet is about building QOS for inter-service-provider networking," Shetty says. "It doesn't talk about how a voice call is connected [when dealing with different providers' networks] or how the video is done."

    Cisco also stresses that it doesn't plan to create a forum similar to Juniper's Infranet Initiative Council; rather, Cisco intends to push its ideas through standards bodies, namely the International Telecommunication Union (ITU) and Alliance for Telecommunications Industry Solutions (ATIS).

  • Finally, as expected, Cisco is announcing a half-sized version of the CRS-1 today (Sources: Cisco Building 'Son of HFR'). The system is an eight-slot chassis, compared with the 16-slot CRS-1, for a throughput of 320 Gbit/s (that's not counting ingress and egress traffic separately). The strategy mimics what Avici and Juniper have done, offering smaller sizes of router to appeal to a wider audience (see Sources: Cisco Building 'Son of HFR', Avici's Incredible Shrinking Router, and Juniper Shrinks Its SuperCore Router).

    The half-sized CRS-1 can be connected in multichassis configurations just as the full CRS-1 can, giving Cisco more options in competing against other multichassis offerings like the Avici TSR or the recently launched Juniper TX Matrix (see Juniper Unveils the TX).

    — Craig Matsumoto, Senior Editor, Light Reading




    SUPERWEBINAR ALERT
    On Thursday, December 9, at 12 noon New York / 9 a.m. California / 5 p.m. London time, during a free hour-long live Web seminar, representatives from Light Reading, EANTC, Agilent, and Cisco will present the results of the Light Reading CRS-1 test and answer questions about them. Telecom Italia will also be participating.

    To register for the December 9 Webinar click here.


  • Page 1 / 7   >   >>
    AAL5 12/5/2012 | 12:59:20 AM
    re: Cisco Scores CRS-1 Customers myhui said "Can you point out details of tests you refer to?",

    The June 2004 Miercom testing. In the multicast/unicast tests OC-192 Unicast performace on the 640 dropped to 60.7%. The 640 also misordered 45% of the OC192 multicast traffic and 15% of the unicast.

    Architecturally the 640 is not suited for multicast applications.

    AAL5
    pxf 12/5/2012 | 12:59:21 AM
    re: Cisco Scores CRS-1 Customers Myhui:

    .............
    The ingress port sends the multicast packet once to the fabric, and magically the egress ports (be they on same slot or different slots) can extract this same packet from the fabric. ....
    ................

    This is no magic, rather a stupid design (though not in its entirity). Since Fabric is a shared resource, What you just created are three congestion points. One at the ingress to Fabric, second at the egress from Fabric to PFE and third the TX itself from the PIC.
    and to boot all this, there is no backpressure mechanism :)

    pxf
    myhui 12/5/2012 | 12:59:24 AM
    re: Cisco Scores CRS-1 Customers AAL5 wrote:

    > the ability for the fabric to duplicate a multicast packet
    > from one Rx card to multiple Tx cards without affecting the
    > throughput of other traffic i.e. unicast traffic.

    That ability is exactly what I claimed: store it as few number of times as possible as the packet travels through the system. On the M series, this was exactly true, in the purest sense. The ingress port sends the multicast packet once to the fabric, and magically the egress ports (be they on same slot or different slots) can extract this same packet from the fabric. Pradeep and others invented the magic part.

    > you can run multicast through a T-640, sending
    multicast
    > traffic to multiple Tx cards without affecting other traffic's
    > throughput?

    Yes. Ingress only needs to send MC packet to fabric once. As for affecting other traffic ... well, if your egress ports are saturated with unicast, then an added load of multicast to those same group of ports will result in dropped packets after the system's queues overflow.

    Can you point out details of tests you refer to?
    AAL5 12/5/2012 | 12:59:35 AM
    re: Cisco Scores CRS-1 Customers myhui wrote "What if multicast traffic was duplicated as close to the output ports as possible, hence reducing the need to store it (or worse, switch it) multiple times in the fabric?"

    What is not in question is the ability to duplicate traffic on the Tx side, but the ability for the fabric to duplicate a multicast packet from one Rx card to multiple Tx cards without affecting the throughput of other traffic i.e. unicast traffic. I was also under the impression that in the T-640 and M-160 multicast was an issue because they are unable to do this.

    Are you claiming this is not the case, i.e. you can run multicast through a T-640, sending multicast traffic to multiple Tx cards without affecting other traffic's throughput? As from testing results I've seen this is not the case.

    There are a number of good things about the 640 but Multicast and Qos at line rate are two of the weaknesses of this system. Also it will be interesting to see how a 40G interface will be implemented if Juniper is going to use their 2*20Gig engine, i.e. how will they avoid the out of order packet problems that plagued the M-160.

    AAL5
    myhui 12/5/2012 | 12:59:40 AM
    re: Cisco Scores CRS-1 Customers pxf wrote:

    > Main issue is that switching fabric does not support
    > replication or prioritization of multicast traffic. Need to
    > use PICs which degrades performance.

    Is that the only basis of your claim: that the switch fabric must support replication of multicast traffic, otherwise it won't qualify as providing good support for multicast? What if multicast traffic was duplicated as close to the output ports as possible, hence reducing the need to store it (or worse, switch it) multiple times in the fabric? It seems your initial assumption is wrong.

    > Actually white paper on T's never talk about switch
    > fabrick thourghtput. All it talks about is a nice marketing
    > term "non blocking fabric" :)

    Why not just run tests on the box, and see whether it handles multicast efficiently? The M and T series handle multicast very, very well. I can assure you that. Just test it. You'll be impressed.
    gotman 12/5/2012 | 12:59:52 AM
    re: Cisco Scores CRS-1 Customers I know, it feels like early 2000 again!! Press releases, product announcement, acquisitions, hiring talent all in one week, after all its the analysts week isn't it? What a better time to dig deep. Wish it was like this across the sector. Its good to leave the year on a high note.
    reoptic 12/5/2012 | 12:59:56 AM
    re: Cisco Scores CRS-1 Customers So you have to give Cisco credit. Invested a ton in CRS1 and it seems to work as Softbank attests, recruited Tony Li back and just bought up BCN...to say nothing of the Procket team they bought. These guys seem to be running the table on new routing technology and talent. With technical leadership and dominant market share...seems promising...only problem is pricing is going through the floor...
    cyberpunk 12/5/2012 | 12:59:56 AM
    re: Cisco Scores CRS-1 Customers Hi,
    Your comment is absolutely correct. The CRS-1 is
    a next-generation platform, running the ENA/IOS-XR
    OS, and is totally modular in its delivery. Plus
    the CRS-1 is a distributed single-to-multi-chassis
    box depending upon the type of configuration you
    choose.
    The CRS-1 is definitely not going to be given away
    free, and it is better than anything either Juniper or Avici has to offer at the current time.
    The only box which came close to it was the PRO-800 which was Procket's box, but the CRS-1 has
    more features and is available in different configurations. The half-QT which is 05 Tbits is
    the low-end (entry-level) core box, followed by
    the medium 1 Tbit, 5 Tbit an 10 Tbit systems which
    are the highe end versions.
    Cp.
    Light-bulb 12/5/2012 | 12:59:57 AM
    re: Cisco Scores CRS-1 Customers Hmm, giving away a CSR in place of a GSR? Hmm, maybe if you only require GE, OC192, and OC768... are there other interfaces for the CRS-1 yet? It's completely a Core box pushing the GSR further to the middle-aggregation...

    Seriously, who ever started this whole thread on giving away CRS-1s is out to lunch.
    marete 12/5/2012 | 12:59:59 AM
    re: Cisco Scores CRS-1 Customers Any news on Axiowave in Marlborough? It's too bad that they had to lay off so many employees. The management team looks strong, and should have been successful like at Nexabit. Anyone know if they developed ASICs for their product?
    Page 1 / 7   >   >>
    HOME
    Sign In
    SEARCH
    CLOSE
    MORE
    CLOSE