lkml.org 
[lkml]   [2000]   [Feb]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRE: Trunking (was: Gigabit Linux Server Bottlenecks)
    -----BEGIN PGP SIGNED MESSAGE-----


    On 08-Feb-2000 Robert Johannes wrote:
    > eh, not sure.
    >
    > I was just thinking of lumping a bunch of interfaces together ( I suppose
    > sun-style) and creating one gigantic pipe from them.

    The following is "as far as I know of" and "supposed to assuming multipath
    forwarding works peroperly". Please correct me if I am wrong.
    All
    #include "stdisclaimer.h"
    apply ;-)

    >
    > Does this need to be supported by the switch or hub for it to work? Can

    Yes, by both switch and host.

    > it also work with a plain-old dual speed hub?

    No, besides one case, see below

    >
    > robert
    >

    There is a piece of junk usually referred as Ethernet stripping. It is what
    MindC**p actually used to win the NT vs Linux benchmark. At least this the way
    I see the data in German's C'T article on a the becnhmark rerun.

    Basically it boils down to the following:

    If connected to a network with more than one interface NT (at least under some
    conditions) does round robin on the Ethers for emitting packets. The Actual IP
    address of the interfaces does not matter. As a result you can get a pretty full
    concatenated pipe if your traffic is:
    A) disbalanced like web or lots of clients pulling stuff off a file server
    B) server is being hit randomly due to DNS round/robin or other load balancing
    technique.

    Note on this one: If you have a look at instructions on using concatenated
    Ethers on NT you will see that "all trunking functions must be turned off".

    In order to get the same result with linux you have to use the following setup:
    1. The service (web, whatever) hanging off a loopback alias on the linux host.
    2. A switch with multipath routing capabilities having a route per each Ether
    to the linux host.
    3. The linux host having multipath enabled (2.2+) and multipath routes back.

    If it does not balance properly in this case there has to be some problem with
    multipath (this is forwarding, so it should work).

    Otherwise it will answer strictly from the interface that has been hit as per
    Mindcraft and C'T benchmarks.

    Brgds,

    - ----------------------------------
    Anton R. Ivanov
    IP Engineer Level3 Communications
    RIPE: ARI2-RIPE E-Mail: Anton Ivanov <aivanov@eu.level3.net>
    @*** Vonnegut's Corollary ***
    Beauty may be only skin deep, but ugliness goes
    right to the core.

    - ----------------------------------
    -----BEGIN PGP SIGNATURE-----
    Version: GnuPG v1.0.0 (GNU/Linux)
    Comment: For info see http://www.gnupg.org

    iQEVAwUBOKKPuylWAw/bM84zAQHocwf+MK2f9evPosfxvBKnWsykuyVeVw2I+8qx
    94cHtFt6+ByYOY1E9EiRpUn8W6vpVgIjvkJA9pRB+P+VkZMq85hcXZZKhb0NqWdf
    OyR/i+gPaeScodjjN99IUpINhD8nwbWK0maRphkQ03UEENILxTEPD4S+SgQDPnwL
    SrjYyEURp33lCKfZP58gB7SQXftaxYReg8IYhIsO6DIa4ujkgZJvTDg/MpPQX7K4
    5imRN8lWHpKdEbQM6ojO+fkxijQGtr0S0oMb1jWCwztXoPSwRN9bhYBxkp+ee39b
    JqJm3Tqr1mUb0bijQyn9oFFwqy58or/P2d+FxqDPS8CePZtkwwCHEg==
    =e3RX
    -----END PGP SIGNATURE-----

    -
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.rutgers.edu
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2005-03-22 13:56    [W:2.170 / U:0.044 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site