Problem - OSPF Unequal Cost Load Balancing (Sort of)

Discussion in 'Cisco' started by NNTP Reader, Jul 22, 2003.

  1. NNTP Reader

    NNTP Reader Guest

    I've got a router that plugs in to an ethernet MAN using both a gigabit
    ethernet interface and a fast ethernet interface with both interfaces
    running OSPF for resiliency.

    Because the GigE interface is the faster one, all outbound traffic is taking
    that interface outbound from the router. (fine)

    The routers at the other end however are MSFC2 cards on our 6500 switches.
    Each interface comes in to a different VLAN but the actual speed (bandwidth
    descriptor) of each VLAN is the same. This is having the effect of load
    balancing the return traffic to the router across both the GigE interface
    and the FastE interface.

    While it is unlikely that the router will ever see more than 200MB of
    sustained traffic, I'd prefer to have the GigE path preferred over the
    100meg path. Fudging the bandwidth descriptor on the VLAN isn't an option
    because there are other routers in the VLANs that have equal cost
    interfaces.

    Any ideas ?

    = K
    NNTP Reader, Jul 22, 2003
    #1
    1. Advertising

  2. NNTP Reader

    Hansang Bae Guest

    In article <>, says...
    > I've got a router that plugs in to an ethernet MAN using both a gigabit
    > ethernet interface and a fast ethernet interface with both interfaces
    > running OSPF for resiliency.
    >
    > Because the GigE interface is the faster one, all outbound traffic is taking
    > that interface outbound from the router. (fine)
    >
    > The routers at the other end however are MSFC2 cards on our 6500 switches.
    > Each interface comes in to a different VLAN but the actual speed (bandwidth
    > descriptor) of each VLAN is the same. This is having the effect of load
    > balancing the return traffic to the router across both the GigE interface
    > and the FastE interface.
    >
    > While it is unlikely that the router will ever see more than 200MB of
    > sustained traffic, I'd prefer to have the GigE path preferred over the
    > 100meg path. Fudging the bandwidth descriptor on the VLAN isn't an option
    > because there are other routers in the VLANs that have equal cost
    > interfaces.



    If you can't use BW statement, then I guess "ip ospf priority" is out
    too? Unlike BGP, you can't make one particular NLRI/route more
    preferable unless you are doing redistribution. It may be ugly, but you
    could redistribute the routes at the 6509 side and mark it as an E1
    route.

    But if they are coming into different VLANs, why can't you use BW or
    ospf cost?

    --

    hsb

    "Somehow I imagined this experience would be more rewarding" Calvin
    *************** USE ROT13 TO SEE MY EMAIL ADDRESS ****************
    ********************************************************************
    Due to the volume of email that I receive, I may not not be able to
    reply to emails sent to my account. Please post a followup instead.
    ********************************************************************
    Hansang Bae, Jul 22, 2003
    #2
    1. Advertising

  3. NNTP Reader

    NNTP Reader Guest

    I can't use the BW descriptor on the other VLAN because that VLAN also
    contains other routers that have interfaces of equal speeds. This is the
    only router where the two interfaces are different speeds. Thus I don't want
    to change the BW descriptor because equal cost load balancing between the 2
    VLAN's is the desired default behavior.

    Ideally, there would be something I could tweak at the interface or ospf
    config level at the 'deviant' router that would let the upstream routers
    know that that path is less desirable.

    = K



    "Hansang Bae" <> wrote in message
    news:...
    > In article <>, says...
    > > I've got a router that plugs in to an ethernet MAN using both a gigabit
    > > ethernet interface and a fast ethernet interface with both interfaces
    > > running OSPF for resiliency.
    > >
    > > Because the GigE interface is the faster one, all outbound traffic is

    taking
    > > that interface outbound from the router. (fine)
    > >
    > > The routers at the other end however are MSFC2 cards on our 6500

    switches.
    > > Each interface comes in to a different VLAN but the actual speed

    (bandwidth
    > > descriptor) of each VLAN is the same. This is having the effect of load
    > > balancing the return traffic to the router across both the GigE

    interface
    > > and the FastE interface.
    > >
    > > While it is unlikely that the router will ever see more than 200MB of
    > > sustained traffic, I'd prefer to have the GigE path preferred over the
    > > 100meg path. Fudging the bandwidth descriptor on the VLAN isn't an

    option
    > > because there are other routers in the VLANs that have equal cost
    > > interfaces.

    >
    >
    > If you can't use BW statement, then I guess "ip ospf priority" is out
    > too? Unlike BGP, you can't make one particular NLRI/route more
    > preferable unless you are doing redistribution. It may be ugly, but you
    > could redistribute the routes at the 6509 side and mark it as an E1
    > route.
    >
    > But if they are coming into different VLANs, why can't you use BW or
    > ospf cost?
    >
    > --
    >
    > hsb
    >
    > "Somehow I imagined this experience would be more rewarding" Calvin
    > *************** USE ROT13 TO SEE MY EMAIL ADDRESS ****************
    > ********************************************************************
    > Due to the volume of email that I receive, I may not not be able to
    > reply to emails sent to my account. Please post a followup instead.
    > ********************************************************************
    NNTP Reader, Jul 22, 2003
    #3
  4. NNTP Reader

    NNTP Reader Guest

    I don't see how adjusting the cost is going to change anything. The router
    with the 2 interfaces already knows that the GigE is faster than the 100meg
    interface and uses the GigE outbound.

    The problem is that the return trip from the MSFCs are load balancing across
    the 2 VLAN's because the VLAN interface speed is the same. It would be nice
    if the MSFC's could determine that the particular neighbor on VLAN B
    established its relationship via a slower interface than the same neighbor
    on VLAN C but this doesn't seem to be the case.

    = K

    "shope" <> wrote in message
    news:itsTa.83$...
    > "NNTP Reader" <> wrote in message
    > news:...
    > > I can't use the BW descriptor on the other VLAN because that VLAN also
    > > contains other routers that have interfaces of equal speeds. This is the
    > > only router where the two interfaces are different speeds. Thus I don't

    > want
    > > to change the BW descriptor because equal cost load balancing between

    the
    > 2
    > > VLAN's is the desired default behavior.
    > >
    > > Ideally, there would be something I could tweak at the interface or ospf
    > > config level at the 'deviant' router that would let the upstream routers
    > > know that that path is less desirable.

    >
    > The OSPF cost on a cisco router is set by default from the bandwidth of

    the
    > interface - but you can set the cost manually on an interface to override
    > that.
    >
    > Seems the simplest way really since it wont have side effects for other
    > things.
    > >
    > > = K
    > >
    > >
    > >
    > > "Hansang Bae" <> wrote in message
    > > news:...
    > > > In article <>, says...
    > > > > I've got a router that plugs in to an ethernet MAN using both a

    > gigabit
    > > > > ethernet interface and a fast ethernet interface with both

    interfaces
    > > > > running OSPF for resiliency.
    > > > >
    > > > > Because the GigE interface is the faster one, all outbound traffic

    is
    > > taking
    > > > > that interface outbound from the router. (fine)
    > > > >
    > > > > The routers at the other end however are MSFC2 cards on our 6500

    > > switches.
    > > > > Each interface comes in to a different VLAN but the actual speed

    > > (bandwidth
    > > > > descriptor) of each VLAN is the same. This is having the effect of

    > load
    > > > > balancing the return traffic to the router across both the GigE

    > > interface
    > > > > and the FastE interface.
    > > > >
    > > > > While it is unlikely that the router will ever see more than 200MB

    of
    > > > > sustained traffic, I'd prefer to have the GigE path preferred over

    the
    > > > > 100meg path. Fudging the bandwidth descriptor on the VLAN isn't an

    > > option
    > > > > because there are other routers in the VLANs that have equal cost
    > > > > interfaces.
    > > >
    > > >
    > > > If you can't use BW statement, then I guess "ip ospf priority" is out
    > > > too? Unlike BGP, you can't make one particular NLRI/route more
    > > > preferable unless you are doing redistribution. It may be ugly, but

    you
    > > > could redistribute the routes at the 6509 side and mark it as an E1
    > > > route.
    > > >
    > > > But if they are coming into different VLANs, why can't you use BW or
    > > > ospf cost?
    > > >
    > > > --
    > > >
    > > > hsb
    > > >
    > > > "Somehow I imagined this experience would be more rewarding" Calvin
    > > > *************** USE ROT13 TO SEE MY EMAIL ADDRESS ****************
    > > > ********************************************************************
    > > > Due to the volume of email that I receive, I may not not be able to
    > > > reply to emails sent to my account. Please post a followup instead.
    > > > ********************************************************************

    > --
    > Regards
    >
    > Stephen Hope - remove xx from email to reply
    >
    >
    NNTP Reader, Jul 23, 2003
    #4
  5. NNTP Reader

    shope Guest

    "NNTP Reader" <> wrote in message
    news:...
    > I don't see how adjusting the cost is going to change anything. The router
    > with the 2 interfaces already knows that the GigE is faster than the

    100meg
    > interface and uses the GigE outbound.


    The router may "know" the speeds of the interfaces or not - MSFCs usually
    talk to VLANs rather than ports.

    If the network was directly connected you would be right - but here it
    learns via OSPF and the OSPF interface costs affect the cost to the
    destination

    So the OSPF costs dictate which routes are preferred - so if you override
    the default cost you can bias the traffic to go out of a specific interface.
    >
    > The problem is that the return trip from the MSFCs are load balancing

    across
    > the 2 VLAN's because the VLAN interface speed is the same. It would be

    nice
    > if the MSFC's could determine that the particular neighbor on VLAN B
    > established its relationship via a slower interface than the same neighbor
    > on VLAN C but this doesn't seem to be the case.


    There is another way you could try - makr the slower interface as "passive"
    in the config, and OSPF wont find the neighbour and build an ajacency, so
    wont learn any routes via that interface.

    I dont think this is what you wanted, since this would prevent you having
    this path as a backup link.
    >
    > = K
    >
    > "shope" <> wrote in message
    > news:itsTa.83$...
    > > "NNTP Reader" <> wrote in message
    > > news:...
    > > > I can't use the BW descriptor on the other VLAN because that VLAN also
    > > > contains other routers that have interfaces of equal speeds. This is

    the
    > > > only router where the two interfaces are different speeds. Thus I

    don't
    > > want
    > > > to change the BW descriptor because equal cost load balancing between

    > the
    > > 2
    > > > VLAN's is the desired default behavior.
    > > >
    > > > Ideally, there would be something I could tweak at the interface or

    ospf
    > > > config level at the 'deviant' router that would let the upstream

    routers
    > > > know that that path is less desirable.

    > >
    > > The OSPF cost on a cisco router is set by default from the bandwidth of

    > the
    > > interface - but you can set the cost manually on an interface to

    override
    > > that.
    > >
    > > Seems the simplest way really since it wont have side effects for other
    > > things.
    > > >
    > > > = K
    > > >
    > > >
    > > >
    > > > "Hansang Bae" <> wrote in message
    > > > news:...
    > > > > In article <>, says...
    > > > > > I've got a router that plugs in to an ethernet MAN using both a

    > > gigabit
    > > > > > ethernet interface and a fast ethernet interface with both

    > interfaces
    > > > > > running OSPF for resiliency.
    > > > > >
    > > > > > Because the GigE interface is the faster one, all outbound traffic

    > is
    > > > taking
    > > > > > that interface outbound from the router. (fine)
    > > > > >
    > > > > > The routers at the other end however are MSFC2 cards on our 6500
    > > > switches.
    > > > > > Each interface comes in to a different VLAN but the actual speed
    > > > (bandwidth
    > > > > > descriptor) of each VLAN is the same. This is having the effect of

    > > load
    > > > > > balancing the return traffic to the router across both the GigE
    > > > interface
    > > > > > and the FastE interface.
    > > > > >
    > > > > > While it is unlikely that the router will ever see more than 200MB

    > of
    > > > > > sustained traffic, I'd prefer to have the GigE path preferred over

    > the
    > > > > > 100meg path. Fudging the bandwidth descriptor on the VLAN isn't an
    > > > option
    > > > > > because there are other routers in the VLANs that have equal cost
    > > > > > interfaces.
    > > > >
    > > > >
    > > > > If you can't use BW statement, then I guess "ip ospf priority" is

    out
    > > > > too? Unlike BGP, you can't make one particular NLRI/route more
    > > > > preferable unless you are doing redistribution. It may be ugly, but

    > you
    > > > > could redistribute the routes at the 6509 side and mark it as an E1
    > > > > route.
    > > > >
    > > > > But if they are coming into different VLANs, why can't you use BW or
    > > > > ospf cost?
    > > > >
    > > > > --
    > > > >
    > > > > hsb
    > > > >
    > > > > "Somehow I imagined this experience would be more rewarding" Calvin
    > > > > *************** USE ROT13 TO SEE MY EMAIL ADDRESS ****************
    > > > > ********************************************************************
    > > > > Due to the volume of email that I receive, I may not not be able to
    > > > > reply to emails sent to my account. Please post a followup instead.
    > > > > ********************************************************************

    > > --
    > > Regards
    > >
    > > Stephen Hope - remove xx from email to reply

    --
    Regards

    Stephen Hope - remove xx from email to reply
    shope, Jul 23, 2003
    #5
  6. NNTP Reader <> wrote in message
    news:...
    > I don't see how adjusting the cost is going to change anything. The router
    > with the 2 interfaces already knows that the GigE is faster than the

    100meg
    > interface and uses the GigE outbound.
    >
    > The problem is that the return trip from the MSFCs are load balancing

    across
    > the 2 VLAN's because the VLAN interface speed is the same. It would be

    nice
    > if the MSFC's could determine that the particular neighbor on VLAN B
    > established its relationship via a slower interface than the same neighbor
    > on VLAN C but this doesn't seem to be the case.


    Consider the outbound traffic from MSFC. If the interface speeds are same
    for that it will make no difference (for the msfc) which interface is being
    used.
    This is how routing works, avoiding secondary effects is not the purpose
    of routing, only the selection of a good route is.

    However, you seem to be trying to solve a secondary problem for avoiding
    overloading the next hop. The next-hop cannot influece the routing of MSFC.
    I think BGP is the only routing protocol with which you can influece how
    the neighbor will route the traffic *to you*. All other protocols make only
    choices for how you route the traffic out of your local node for a shortest
    possible route in you local nodes point of view.

    OSPF cost can be used as a static solution. Add a worse (or better) cost
    for one VLAN on both of your MSFCs and they will prefer the vlan with
    better cost. (Pain to administer and maintain if you have lots of nodes.)

    However, this leads to another problem: all the traffic is routed through
    the
    better vlan now and the other vlan is idling with no traffic. I guess from
    the
    previous that this means routing all through a single 100Mbit interface out
    of the MSFC. This might or might not be acceptable.

    Simple solution for unequal load balancing with OSPF:

    Have vlan A & B on a gig-interface and VLAN C on the 100M interface.
    If outbound costs are equal this will mean equal load balancing for vlans
    A,B,C. Of this total traffic 2/3 will be on the gigE and 1/3 on the
    FastEthernet
    because the VLAN distribution.

    This means adding more vlans but can be used for a sort of unequal load
    balancing which works with complete support of the routeing protocol.
    All you need to do is to engineer suitable number of neighboring interfaces.
    --
    Harri
    Harri Suomalainen, Jul 24, 2003
    #6
  7. NNTP Reader <> wrote in message
    news:...
    > I don't see how adjusting the cost is going to change anything. The router
    > with the 2 interfaces already knows that the GigE is faster than the

    100meg
    > interface and uses the GigE outbound.
    >
    > The problem is that the return trip from the MSFCs are load balancing

    across
    > the 2 VLAN's because the VLAN interface speed is the same. It would be

    nice
    > if the MSFC's could determine that the particular neighbor on VLAN B
    > established its relationship via a slower interface than the same neighbor
    > on VLAN C but this doesn't seem to be the case.


    Consider the outbound traffic from MSFC. If the interface speeds are same
    for that it will make no difference (for the msfc) which interface is being
    used.
    This is how routing works, avoiding secondary effects is not the purpose
    of routing, only the selection of a good route is.

    However, you seem to be trying to solve a secondary problem for avoiding
    overloading the next hop. The next-hop cannot influece the routing of MSFC.
    I think BGP is the only routing protocol with which you can influece how
    the neighbor will route the traffic *to you*. All other protocols make only
    choices for how you route the traffic out of your local node for a shortest
    possible route in you local nodes point of view.

    OSPF cost can be used as a static solution. Add a worse (or better) cost
    for one VLAN on both of your MSFCs and they will prefer the vlan with
    better cost. (Pain to administer and maintain if you have lots of nodes.)

    However, this leads to another problem: all the traffic is routed through
    the
    better vlan now and the other vlan is idling with no traffic. I guess from
    the
    previous that this means routing all through a single 100Mbit interface out
    of the MSFC. This might or might not be acceptable.

    Simple solution for unequal load balancing with OSPF:

    Have vlan A & B on a gig-interface and VLAN C on the 100M interface.
    If outbound costs are equal this will mean equal load balancing for vlans
    A,B,C. Of this total traffic 2/3 will be on the gigE and 1/3 on the
    FastEthernet
    because the VLAN distribution.

    This means adding more vlans but can be used for a sort of unequal load
    balancing which works with complete support of the routeing protocol.
    All you need to do is to engineer suitable number of neighboring interfaces.
    --
    Harri
    Harri Suomalainen, Jul 24, 2003
    #7
  8. NNTP Reader

    newsreader Guest

    How about EIGRP with variance? If you're all Cisco, perhaps you could try
    that...

    "NNTP Reader" <> wrote in message
    news:...
    > I've got a router that plugs in to an ethernet MAN using both a gigabit
    > ethernet interface and a fast ethernet interface with both interfaces
    > running OSPF for resiliency.
    >
    > Because the GigE interface is the faster one, all outbound traffic is

    taking
    > that interface outbound from the router. (fine)
    >
    > The routers at the other end however are MSFC2 cards on our 6500 switches.
    > Each interface comes in to a different VLAN but the actual speed

    (bandwidth
    > descriptor) of each VLAN is the same. This is having the effect of load
    > balancing the return traffic to the router across both the GigE interface
    > and the FastE interface.
    >
    > While it is unlikely that the router will ever see more than 200MB of
    > sustained traffic, I'd prefer to have the GigE path preferred over the
    > 100meg path. Fudging the bandwidth descriptor on the VLAN isn't an option
    > because there are other routers in the VLANs that have equal cost
    > interfaces.
    >
    > Any ideas ?
    >
    > = K
    >
    >



    ---
    Outgoing mail is certified Virus Free.
    Checked by AVG anti-virus system (http://www.grisoft.com).
    Version: 6.0.502 / Virus Database: 300 - Release Date: 7/18/2003
    newsreader, Jul 25, 2003
    #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. Aaron Woody
    Replies:
    0
    Views:
    873
    Aaron Woody
    Oct 20, 2003
  2. Rob
    Replies:
    7
    Views:
    4,522
    Terry Baranski
    Feb 21, 2004
  3. Lior Shadhan

    Unequal Load Balancing + GLBP

    Lior Shadhan, Jan 5, 2005, in forum: Cisco
    Replies:
    0
    Views:
    769
    Lior Shadhan
    Jan 5, 2005
  4. jimbo
    Replies:
    1
    Views:
    1,048
    John Agosta
    Apr 2, 2005
  5. Replies:
    7
    Views:
    4,820
    Charlie Root
    Feb 23, 2006
Loading...

Share This Page