How many T1's can a 2621 handle?

Discussion in 'Cisco' started by WAState, Feb 21, 2005.

  1. WAState

    WAState Guest

    What is the practical limit on the number of T1's that a 2621 (the
    original, NOT the XM version) can handle before you run out of CPU?

    By associating multiple T1's via MultiLink, I'm wondering how large a
    WAN pipe the 2621 can handle. The device can *physically* accommodate
    up to four WIC-1DSU-T1's: Two in its native WAN slots, and two more in
    an optional NM-2W. So theoretically you could have four T1's, or ~6Mbps
    in each direction (12Mbps total), in a MultiLink setup.

    One reference on Cisco's website says the 2621 can handle ~8Mbps, but
    1) I've always understood that the 2621 can sustain full FE wire
    speeds, which is quite a bit more than 8Mbps; and 2) I don't know if
    that 8Mbps figure means one side of a link or both sides. T1's are
    commonly referred to as 1.544Mbps, but they are full duplex so the
    total traffic is really ~3Mbps... I don't know if Cisco's "8Mbps"
    figure means unidirectional or bidirectional traffic.

    Anyone ever put three or four WIC's in a single 2621? Did it work OK?
    WAState, Feb 21, 2005
    #1
    1. Advertising

  2. Hello, WAState!
    You wrote on 21 Feb 2005 10:39:11 -0800:

    W> What is the practical limit on the number of T1's that a 2621 (the
    W> original, NOT the XM version) can handle before you run out of
    W> CPU?

    W> By associating multiple T1's via MultiLink, I'm wondering how
    W> large a WAN pipe the 2621 can handle. The device can *physically*
    W> accommodate up to four WIC-1DSU-T1's: Two in its native WAN slots,
    W> and two more in an optional NM-2W. So theoretically you could have
    W> four T1's, or ~6Mbps in each direction (12Mbps total), in a
    W> MultiLink setup.

    Well, physically you can have 8 T1. You can use WICs with 2 serial port each and
    external CSU/DSU. I didn't bother to check what the actual limitation on number
    of serial ports/WICs you can have in 2621.

    W> One reference on Cisco's website says the 2621 can handle ~8Mbps,
    W> but 1) I've always understood that the 2621 can sustain full FE
    W> wire speeds, which is quite a bit more than 8Mbps;

    Never ever. According to

    http://www.cisco.com/warp/public/765/tools/quickreference/routerperformance.pdf

    2621 can handle up to 12.80 Mbps. I don't know which packet size they were using
    to get this number. One friend of my claimed that he couldn't get more than 20
    Mbps from one FE to another FE with 1500 packets. I didn't verify either number.

    W> and 2) I don't know if that 8Mbps figure means one side of a link or both
    W> sides. T1's are commonly referred to as 1.544Mbps, but they are full
    W> duplex so the total traffic is really ~3Mbps... I don't know if
    W> Cisco's "8Mbps" figure means unidirectional or bidirectional
    W> traffic.

    In this case I think it would be 4 T1. 6 Mbps one way and 6 another. Chances are
    that 12.80 Mbps was squeezed from router without any additional processing like
    NAT, ACL, PBR, etc. To be on a safe side I would say two T1 per 2621 looks like
    a maximum. But if you T1 is not heavily utilized you might get away with more
    than 2.

    With best regards,
    Andrey.
    Andrey Tarasov, Feb 21, 2005
    #2
    1. Advertising

  3. WAState

    WAState Guest

    Thanks for the comments.

    As an aside, I've been playing with throughput on 2600's lately. I seem
    to get the same throughput no matter whether the packets are all
    process-switched or if I enable CEF+NetFlow with flow acceleration. CPU
    loading is about the same during the tests, too. I must say I was
    rather surprised by those results, and my testing continues.
    WAState, Feb 22, 2005
    #3
  4. WAState

    Guest

    At a guess there is something wrong with your tests.

    I have done tests on cisco routers and saw the 'expected' performance
    variations and CPU loading changes. I reckon on a factor of 10 (or
    more) difference between and fast switching and process switching.

    MPPP is CPU intensive and the performance numbers with MPPP are not in
    any way related to any general published performance figured. For n
    links each packet is divided into n pieces and each piece sent down a
    different link.

    MPPP over 30 channels of 64k each on a European PRI sucked up a100% CPU
    on a 4700M. (100MHz or was it 150MHz? R4700).
    , Feb 23, 2005
    #4
  5. Hello, !
    You wrote on 23 Feb 2005 03:24:45 -0800:

    a> MPPP is CPU intensive and the performance numbers with MPPP are
    a> not in any way related to any general published performance
    a> figured. For n links each packet is divided into n pieces and each
    a> piece sent down a different link.

    It depends. What I seen on 2600 series, if you are not using interleaving or if
    links in MPPP bundle are fast enough, there would be no fragmentation
    whatsoever.

    With best regards,
    Andrey.
    Andrey Tarasov, Feb 23, 2005
    #5
  6. WAState

    AnyBody43 Guest

    "Andrey Tarasov" <> wrote in message news:<cviebj$1th9$>...
    > Hello, !
    > You wrote on 23 Feb 2005 03:24:45 -0800:
    >
    > a> MPPP is CPU intensive and the performance numbers with MPPP are
    > a> not in any way related to any general published performance
    > a> figured. For n links each packet is divided into n pieces and each
    > a> piece sent down a different link.
    >
    > It depends. What I seen on 2600 series, if you are not using interleaving or if
    > links in MPPP bundle are fast enough, there would be no fragmentation
    > whatsoever.
    >
    > With best regards,
    > Andrey.


    Andrey,
    I have only just noticed your response.

    I believe that _by-definition_ MPPP divides the packets
    such that in the case of N links there are N fragments.

    However in general with respect to MPPP performance
    I have no recent experience of mppp and may be out
    of date. Clearly it would be perfectly possible to
    fragment the packets in hardware and incur
    no performance penalty whatsoever. It seems
    unlikely to me that a 2621 does this.

    A key benefit of fragmenting the packets in this way
    is that the Transmission Delay is reduced. Additionally
    packet order is conserved which is not the case with
    some other load sharing techniques.
    AnyBody43, Mar 19, 2005
    #6
  7. Hello, AnyBody43!
    You wrote on 19 Mar 2005 05:32:20 -0800:

    ??>> It depends. What I seen on 2600 series, if you are not using
    ??>> interleaving or if links in MPPP bundle are fast enough, there
    ??>> would be no fragmentation whatsoever.

    A> I believe that _by-definition_ MPPP divides the packets
    A> such that in the case of N links there are N fragments.

    From RFC 1990 -
    ....
    Systems implementing the multilink procedure are not required to fragment small
    packets. There is also no requirement that the segments be of equal sizes, or
    that packets must be broken up at all.
    ....

    A> However in general with respect to MPPP performance
    A> I have no recent experience of mppp and may be out
    A> of date. Clearly it would be perfectly possible to
    A> fragment the packets in hardware and incur
    A> no performance penalty whatsoever.

    And that is normally found on IMA devices.

    A> A key benefit of fragmenting the packets in this way
    A> is that the Transmission Delay is reduced. Additionally
    A> packet order is conserved which is not the case with
    A> some other load sharing techniques.

    Serialization delay on T1 for 1500 bytes is less than 8ms. Packet order is
    preserved even though packets are not fragmented. There is a special MLPPP
    buffer dealing with packets re-assembly and re-ordering.

    With best regards,
    Andrey.
    Andrey Tarasov, Mar 19, 2005
    #7
  8. WAState

    Guest

    Thanks.

    I did consider heading for the RFCs but was pretty convinced.
    Wrong, again!

    Thanks a lot.
    , Mar 19, 2005
    #8
    1. Advertising

Want to reply to this thread or ask your own question?

It takes just 2 minutes to sign up (and it's free!). Just click the sign up button to choose a username and then you can ask your own questions on the forum.
Similar Threads
  1. =?Utf-8?B?TmlnZWwgQXN0b24=?=

    Can VSS handle Unicode files?

    =?Utf-8?B?TmlnZWwgQXN0b24=?=, May 19, 2004, in forum: Microsoft Certification
    Replies:
    0
    Views:
    598
    =?Utf-8?B?TmlnZWwgQXN0b24=?=
    May 19, 2004
  2. Replies:
    3
    Views:
    1,185
  3. Josh
    Replies:
    2
    Views:
    431
    Michael Heitland
    Feb 16, 2004
  4. jlm33990
    Replies:
    4
    Views:
    2,811
    stephan
    May 4, 2006
  5. fred do

    Can a 300W PSU handle this baby?

    fred do, Jun 24, 2003, in forum: Computer Support
    Replies:
    14
    Views:
    761
    Greg M
    Jun 26, 2003
Loading...

Share This Page