[j-nsp] BGP and OSPF ECMP
Harry Reynolds
harry at juniper.net
Fri Jul 18 14:58:29 EDT 2008
If you are sending to multiple destinations, then by default per prefix lb kicks in. If we are talking PE router then I believe its choice of p-facing NH is based on ce facing hash (!mpls), but for p-router yes, it will be a mpls label stack hash, but for a single vrf with a single pe-ce facing interface I'd expect the same label stack for all routes.
Regards
> -----Original Message-----
> From: Marlon Duksa [mailto:mduksa at gmail.com]
> Sent: Friday, July 18, 2008 11:43 AM
> To: Harry Reynolds
> Cc: juniper-nsp at puck.nether.net
> Subject: Re: [j-nsp] BGP and OSPF ECMP
>
> I'm sending traffic to 50K IP destinations. All this is in
> VPN (2547). So I should have 50K flows if destination IP is
> taken into account during hashing. Please keep in mind that
> this IP is encapsulated in two MPLS headers as the traffic is
> going through MPLS tunnel (LDP based) and then the second
> label is the vrf label. The next hop (PE) router is
> advertised through OSPF.
> Thanks,
> Marlon
>
>
>
>
>
> On Thu, Jul 17, 2008 at 5:24 PM, Harry Reynolds
> <harry at juniper.net> wrote:
>
>
> Hmm. Normally for LB we expect to see a ulst next hop,
> which in turn contains two or more ucst NHs. The FT display
> indicates that a single NH is installed for that prefix.
>
> Can you describe the nature of the test traffic? If its
> all one flow I'd expect all to take the same NH even when LB
> is in effect. AFAIK, MX behavior is no different here.
>
>
>
> Regards
>
>
>
>
>
>
>
> > -----Original Message-----
> > From: Marlon Duksa [mailto:mduksa at gmail.com]
> > Sent: Thursday, July 17, 2008 1:57 PM
> > To: Harry Reynolds; juniper-nsp at puck.nether.net
> > Subject: Re: [j-nsp] BGP and OSPF ECMP
> >
> > I tried this on MX960 and it seams that load balancing is on
> > by default. How do I turn load balancing off?
> > In the RT I can see both next hops:
> > 2.2.2.2/32 *[OSPF/10] 02:21:29, metric 1
> > to 10.0.2.2 via ge-8/0/5.0
> > > to 10.0.3.2 via ge-5/1/0.0
> >
> >
> > In the FT I can see only one path:
> > 2.2.2.2/32 user 1 10.0.3.2 ucst 572
> > 5 ge-5/1/0.0
> >
> > But when I actually send traffic and run 'monitor interface
> > traffic', I see that traffic is going out over both links.
> > This 2.2.2.2 next hop is actually a PE router in VPN.
> >
> > Is load balancing on by default on MX? I didn't configure
> > anything special to turn it on.
> >
> >
> > On Tue, Jul 15, 2008 at 6:43 AM, Boyd, Benjamin R
> > <Benjamin.R.Boyd at windstream.com> wrote:
> >
> >
> > You can further control the flow with:
> >
> > [edit forwarding-options hash-key]
> >
> > family inet {
> > layer-3;
> > layer-4;
> > }
> > family mpls {
> > label-1;
> > label-2;
> > }
> >
> > If you're load balancing over a decent amount of hops I
> > wouldn't hash
> > both layer-3/layer-4 and label-1/label-2 on every
> > router because then
> > your per-flow becomes quite segregated and you might
> > find issues where
> > your load balancing on many routers will break the
> > seemingly load
> > balancing of another router since the hashing algorithm
> > will produce the
> > same result on each router.
> >
> > -Ben
> >
> >
> > -----Original Message-----
> > From: juniper-nsp-bounces at puck.nether.net
> >
> > [mailto:juniper-nsp-bounces at puck.nether.net] On Behalf
> > Of Harry Reynolds
> > Sent: Monday, July 14, 2008 6:18 PM
> > To: Marlon Duksa; juniper-nsp at puck.nether.net
> > Subject: Re: [j-nsp] BGP and OSPF ECMP
> >
> > Do you have a per-packet (really per flow) LB policy
> > applied to the
> > forwarding table?
> >
> > This is needed to install two or more forwarding paths
> > in the pfe. BGP
> > multipath is a control plane tie breaker, but by
> > default only one of the
> > winners is installed as active in the FT.
> >
> > HTHs
> >
> >
> > [edit]
> > regress at asahi# show policy-options
> > policy-statement lb {
> > then {
> > load-balance per-packet;
> > }
> > }
> >
> > [edit]
> > regress at asahi# show routing-options forwarding-table
> > traceoptions {
> > file forwarding_table;
> > flag route detail;
> > }
> > export lb;
> >
> >
> > > -----Original Message-----
> > > From: juniper-nsp-bounces at puck.nether.net
> > > [mailto:juniper-nsp-bounces at puck.nether.net] On
> > Behalf Of Marlon Duksa
> > > Sent: Monday, July 14, 2008 3:22 PM
> > > To: juniper-nsp at puck.nether.net
> > > Subject: [j-nsp] BGP and OSPF ECMP
> > >
> > > Does anyone know how to enable ecmp in Junos
> (I'm on M320
> > > with Junos 9.0)?
> > >
> > > I have a vrf with BGP configured for multipath. In the
> > > forwarding table I see only one path
> installed even though I
> > > have two equal cost physical links to the NH.
> > > My IGP is OSPF and in the routing table OSPF
> is showing two
> > > paths (only one active >) and in the fwd
> table only one path
> > > is selected. How do I force Junos to select
> both links.
> > >
> > > Similar with LDP which is used for the
> transport (tunnel) in
> > > this VPN environment.
> > >
> > > So it looks to me that multipath in BGP is
> not taking effect
> > > because underlying protocols (OSPF and LDP)
> is not utilizing
> > > both links??
> > > Thanks,
> > > Marlon
> > > _______________________________________________
> > > juniper-nsp mailing list juniper-nsp at puck.nether.net
> > > https://puck.nether.net/mailman/listinfo/juniper-nsp
> > >
> > _______________________________________________
> > juniper-nsp mailing list juniper-nsp at puck.nether.net
> > https://puck.nether.net/mailman/listinfo/juniper-nsp
> >
> >
> >
> >
> >
> > **************************************************************
> > *************************
> >
> > The information contained in this message, including
> > attachments, may contain
> > privileged or confidential information that is intended
> > to be delivered only to the
> > person identified above. If you are not the intended
> > recipient, or the person
> > responsible for delivering this message to the intended
> > recipient, Windstream requests
> > that you immediately notify the sender and asks that
> > you do not read the message or its
> > attachments, and that you delete them without copying
> > or sending them to anyone else.
> >
> >
> >
> >
> >
>
>
>
>
More information about the juniper-nsp
mailing list