Mailing List Archive

Since the IP2 processor, load balancing per packet is a bit of
a misnomer. A modern Juniper will load balance per *flow*,
even though the config says per-packet.

You cannot do true per packet load balancing on a Juniper. Which
is a good thing at the datarates they're designed for.

So if you're measuring with a single flow, this might explain it.

BB

-----Original Message-----
From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
Sent: maandag 25 november 2002 16:59
To: juniper-nsp@puck.nether.net
Subject: [j-nsp] per packet/flow load balancing.


Hi,

This has been discussed many times here. Apologize if someone think this is
too simple. But only now I give it a try and without luck.

I have setup this in the test lab:

/-- IBGP -- router A -- EBGP --\
router C < > router D
\-- IBGP -- router B -- EBGP --/

I believe there are equal cost paths but the router C doesn't seem to be
able the load balance per packet.

Below is a snapshot from router C:

192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
192.168.9.2
AS path: 64665 64665 I
to 192.168.4.1 via ge-0/1/0.0
> to 192.168.4.2 via ge-0/1/0.0
192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
192.168.9.2
AS path: 64665 64665 I
> to 192.168.4.1 via ge-0/1/0.0
to 192.168.4.2 via ge-0/1/0.0

router C# show policy-options policy-statement load-balancing-policy
from as-path TEST-AS;
then {
load-balance per-packet;
}

router C# show policy-options as-path TEST-AS
"64665 .*";

router C# show routing-options forwarding-table
export load-balancing-policy;

I have even configured "multipath" at BGP but doesn't help still.

I wonder how do I verify whether the per packet load balancing is being
enabled other than traceroute.

Thanks!

Jonathan.

_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
http://puck.nether.net/mailman/listinfo/juniper-nsp

So that means the "load-balance per-packet" is useless?


----- Original Message -----
From: "Ben Buxton" <b.buxton@planettechnologies.nl>
To: "'Jonathan Tse'" <jonathantse@pacific.net.sg>;
<juniper-nsp@puck.nether.net>
Sent: Tuesday, November 26, 2002 12:22 AM
Subject: RE: [j-nsp] per packet/flow load balancing.


>
> Since the IP2 processor, load balancing per packet is a bit of
> a misnomer. A modern Juniper will load balance per *flow*,
> even though the config says per-packet.
>
> You cannot do true per packet load balancing on a Juniper. Which
> is a good thing at the datarates they're designed for.
>
> So if you're measuring with a single flow, this might explain it.
>
> BB
>
> -----Original Message-----
> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> Sent: maandag 25 november 2002 16:59
> To: juniper-nsp@puck.nether.net
> Subject: [j-nsp] per packet/flow load balancing.
>
>
> Hi,
>
> This has been discussed many times here. Apologize if someone think this
is
> too simple. But only now I give it a try and without luck.
>
> I have setup this in the test lab:
>
> /-- IBGP -- router A -- EBGP --\
> router C < > router D
> \-- IBGP -- router B -- EBGP --/
>
> I believe there are equal cost paths but the router C doesn't seem to be
> able the load balance per packet.
>
> Below is a snapshot from router C:
>
> 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> to 192.168.4.1 via ge-0/1/0.0
> > to 192.168.4.2 via ge-0/1/0.0
> 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> > to 192.168.4.1 via ge-0/1/0.0
> to 192.168.4.2 via ge-0/1/0.0
>
> router C# show policy-options policy-statement load-balancing-policy
> from as-path TEST-AS;
> then {
> load-balance per-packet;
> }
>
> router C# show policy-options as-path TEST-AS
> "64665 .*";
>
> router C# show routing-options forwarding-table
> export load-balancing-policy;
>
> I have even configured "multipath" at BGP but doesn't help still.
>
> I wonder how do I verify whether the per packet load balancing is being
> enabled other than traceroute.
>
> Thanks!
>
> Jonathan.
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>

No, enabling this flag on the IP2 will load balance per flow. The name
is a misnomer but it has very real effects.

-- steve

-----Original Message-----
From: juniper-nsp-admin@puck.nether.net
[mailto:juniper-nsp-admin@puck.nether.net] On Behalf Of Jonathan Tse
Sent: Monday, November 25, 2002 10:28 AM
To: Ben Buxton; juniper-nsp@puck.nether.net
Subject: Re: [j-nsp] per packet/flow load balancing.

So that means the "load-balance per-packet" is useless?


----- Original Message -----
From: "Ben Buxton" <b.buxton@planettechnologies.nl>
To: "'Jonathan Tse'" <jonathantse@pacific.net.sg>;
<juniper-nsp@puck.nether.net>
Sent: Tuesday, November 26, 2002 12:22 AM
Subject: RE: [j-nsp] per packet/flow load balancing.


>
> Since the IP2 processor, load balancing per packet is a bit of
> a misnomer. A modern Juniper will load balance per *flow*,
> even though the config says per-packet.
>
> You cannot do true per packet load balancing on a Juniper. Which
> is a good thing at the datarates they're designed for.
>
> So if you're measuring with a single flow, this might explain it.
>
> BB
>
> -----Original Message-----
> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> Sent: maandag 25 november 2002 16:59
> To: juniper-nsp@puck.nether.net
> Subject: [j-nsp] per packet/flow load balancing.
>
>
> Hi,
>
> This has been discussed many times here. Apologize if someone think
this
is
> too simple. But only now I give it a try and without luck.
>
> I have setup this in the test lab:
>
> /-- IBGP -- router A -- EBGP --\
> router C < > router D
> \-- IBGP -- router B -- EBGP --/
>
> I believe there are equal cost paths but the router C doesn't seem to
be
> able the load balance per packet.
>
> Below is a snapshot from router C:
>
> 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> to 192.168.4.1 via ge-0/1/0.0
> > to 192.168.4.2 via ge-0/1/0.0
> 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> > to 192.168.4.1 via ge-0/1/0.0
> to 192.168.4.2 via ge-0/1/0.0
>
> router C# show policy-options policy-statement load-balancing-policy
> from as-path TEST-AS;
> then {
> load-balance per-packet;
> }
>
> router C# show policy-options as-path TEST-AS
> "64665 .*";
>
> router C# show routing-options forwarding-table
> export load-balancing-policy;
>
> I have even configured "multipath" at BGP but doesn't help still.
>
> I wonder how do I verify whether the per packet load balancing is
being
> enabled other than traceroute.
>
> Thanks!
>
> Jonathan.
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>

_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
http://puck.nether.net/mailman/listinfo/juniper-nsp

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Not entirely. Load balance per packet causes per flow load balancing. If
you don't specify it, you get per destination load balancing.

Guy

> -----Original Message-----
> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> Sent: Monday, November 25, 2002 4:28 PM
> To: Ben Buxton; juniper-nsp@puck.nether.net
> Subject: Re: [j-nsp] per packet/flow load balancing.
>
>
> So that means the "load-balance per-packet" is useless?
>
>
> ----- Original Message -----
> From: "Ben Buxton" <b.buxton@planettechnologies.nl>
> To: "'Jonathan Tse'" <jonathantse@pacific.net.sg>;
> <juniper-nsp@puck.nether.net>
> Sent: Tuesday, November 26, 2002 12:22 AM
> Subject: RE: [j-nsp] per packet/flow load balancing.
>
>
> >
> > Since the IP2 processor, load balancing per packet is a bit of
> > a misnomer. A modern Juniper will load balance per *flow*,
> > even though the config says per-packet.
> >
> > You cannot do true per packet load balancing on a Juniper. Which
> > is a good thing at the datarates they're designed for.
> >
> > So if you're measuring with a single flow, this might explain it.
> >
> > BB
> >
> > -----Original Message-----
> > From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> > Sent: maandag 25 november 2002 16:59
> > To: juniper-nsp@puck.nether.net
> > Subject: [j-nsp] per packet/flow load balancing.
> >
> >
> > Hi,
> >
> > This has been discussed many times here. Apologize if
> someone think this
> is
> > too simple. But only now I give it a try and without luck.
> >
> > I have setup this in the test lab:
> >
> > /-- IBGP -- router A -- EBGP --\
> > router C < > router D
> > \-- IBGP -- router B -- EBGP --/
> >
> > I believe there are equal cost paths but the router C
> doesn't seem to be
> > able the load balance per packet.
> >
> > Below is a snapshot from router C:
> >
> > 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> > 192.168.9.2
> > AS path: 64665 64665 I
> > to 192.168.4.1 via ge-0/1/0.0
> > > to 192.168.4.2 via ge-0/1/0.0
> > 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> > 192.168.9.2
> > AS path: 64665 64665 I
> > > to 192.168.4.1 via ge-0/1/0.0
> > to 192.168.4.2 via ge-0/1/0.0
> >
> > router C# show policy-options policy-statement load-balancing-policy
> > from as-path TEST-AS;
> > then {
> > load-balance per-packet;
> > }
> >
> > router C# show policy-options as-path TEST-AS
> > "64665 .*";
> >
> > router C# show routing-options forwarding-table
> > export load-balancing-policy;
> >
> > I have even configured "multipath" at BGP but doesn't help still.
> >
> > I wonder how do I verify whether the per packet load
> balancing is being
> > enabled other than traceroute.
> >
> > Thanks!
> >
> > Jonathan.
> >
> > _______________________________________________
> > juniper-nsp mailing list juniper-nsp@puck.nether.net
> > http://puck.nether.net/mailman/listinfo/juniper-nsp
> >
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>

-----BEGIN PGP SIGNATURE-----
Version: PGP 8.0 (Build 349) Beta

iQA/AwUBPeJRX43dwu/Ss2PCEQID1gCghv5xIK8QJ4Tx49ME4U+0N7vt914AoMoL
lEHYa5ljdvZNeaH1MnFk8PCJ
=cAZf
-----END PGP SIGNATURE-----


This e-mail is private and may be confidential and is for the intended
recipient only. If misdirected, please notify us by telephone and confirm
that it has been deleted from your system and any copies destroyed. If you
are not the intended recipient you are strictly prohibited from using,
printing, copying, distributing or disseminating this e-mail or any
information contained in it. We use reasonable endeavors to virus scan all
e-mails leaving the Company but no warranty is given that this e-mail and
any attachments are virus free. You should undertake your own virus
checking. The right to monitor e-mail communications through our network is
reserved by us.

Jonathan,

If you're trying to balance a *single* flow across multiple links, "load-balance per-packet" will not do it. If your trying to balance *multiple* flows across multiple links, "load-balance per-packet" will do it.

Andy

>-----Original Message-----
>From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
>Sent: Monday, November 25, 2002 11:28 AM
>To: Ben Buxton; juniper-nsp@puck.nether.net
>Subject: Re: [j-nsp] per packet/flow load balancing.
>
>
>So that means the "load-balance per-packet" is useless?
>
>
>----- Original Message -----
>From: "Ben Buxton" <b.buxton@planettechnologies.nl>
>To: "'Jonathan Tse'" <jonathantse@pacific.net.sg>;
><juniper-nsp@puck.nether.net>
>Sent: Tuesday, November 26, 2002 12:22 AM
>Subject: RE: [j-nsp] per packet/flow load balancing.
>
>
>>
>> Since the IP2 processor, load balancing per packet is a bit of
>> a misnomer. A modern Juniper will load balance per *flow*,
>> even though the config says per-packet.
>>
>> You cannot do true per packet load balancing on a Juniper. Which
>> is a good thing at the datarates they're designed for.
>>
>> So if you're measuring with a single flow, this might explain it.
>>
>> BB
>>
>> -----Original Message-----
>> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
>> Sent: maandag 25 november 2002 16:59
>> To: juniper-nsp@puck.nether.net
>> Subject: [j-nsp] per packet/flow load balancing.
>>
>>
>> Hi,
>>
>> This has been discussed many times here. Apologize if
>someone think this
>is
>> too simple. But only now I give it a try and without luck.
>>
>> I have setup this in the test lab:
>>
>> /-- IBGP -- router A -- EBGP --\
>> router C < > router D
>> \-- IBGP -- router B -- EBGP --/
>>
>> I believe there are equal cost paths but the router C
>doesn't seem to be
>> able the load balance per packet.
>>
>> Below is a snapshot from router C:
>>
>> 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
>> 192.168.9.2
>> AS path: 64665 64665 I
>> to 192.168.4.1 via ge-0/1/0.0
>> > to 192.168.4.2 via ge-0/1/0.0
>> 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
>> 192.168.9.2
>> AS path: 64665 64665 I
>> > to 192.168.4.1 via ge-0/1/0.0
>> to 192.168.4.2 via ge-0/1/0.0
>>
>> router C# show policy-options policy-statement load-balancing-policy
>> from as-path TEST-AS;
>> then {
>> load-balance per-packet;
>> }
>>
>> router C# show policy-options as-path TEST-AS
>> "64665 .*";
>>
>> router C# show routing-options forwarding-table
>> export load-balancing-policy;
>>
>> I have even configured "multipath" at BGP but doesn't help still.
>>
>> I wonder how do I verify whether the per packet load
>balancing is being
>> enabled other than traceroute.
>>
>> Thanks!
>>
>> Jonathan.
>>
>> _______________________________________________
>> juniper-nsp mailing list juniper-nsp@puck.nether.net
>> http://puck.nether.net/mailman/listinfo/juniper-nsp
>>
>
>_______________________________________________
>juniper-nsp mailing list juniper-nsp@puck.nether.net
>http://puck.nether.net/mailman/listinfo/juniper-nsp
>

"load-balance per-packet" will allow you to load balance traffic across multiple equal-cost paths. However, this is not done on a per packet basis. It is done on a per flow basis. Keeping the flows together is a good thing. You will need to setup multiple streams in your scenario below to accomplish what you want. Thanks.

T
-----Original Message-----
From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
Sent: Monday, November 25, 2002 8:28 AM
To: Ben Buxton; juniper-nsp@puck.nether.net
Subject: Re: [j-nsp] per packet/flow load balancing.


So that means the "load-balance per-packet" is useless?


----- Original Message -----
From: "Ben Buxton" <b.buxton@planettechnologies.nl>
To: "'Jonathan Tse'" <jonathantse@pacific.net.sg>;
<juniper-nsp@puck.nether.net>
Sent: Tuesday, November 26, 2002 12:22 AM
Subject: RE: [j-nsp] per packet/flow load balancing.


>
> Since the IP2 processor, load balancing per packet is a bit of
> a misnomer. A modern Juniper will load balance per *flow*,
> even though the config says per-packet.
>
> You cannot do true per packet load balancing on a Juniper. Which
> is a good thing at the datarates they're designed for.
>
> So if you're measuring with a single flow, this might explain it.
>
> BB
>
> -----Original Message-----
> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> Sent: maandag 25 november 2002 16:59
> To: juniper-nsp@puck.nether.net
> Subject: [j-nsp] per packet/flow load balancing.
>
>
> Hi,
>
> This has been discussed many times here. Apologize if someone think this
is
> too simple. But only now I give it a try and without luck.
>
> I have setup this in the test lab:
>
> /-- IBGP -- router A -- EBGP --\
> router C < > router D
> \-- IBGP -- router B -- EBGP --/
>
> I believe there are equal cost paths but the router C doesn't seem to be
> able the load balance per packet.
>
> Below is a snapshot from router C:
>
> 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> to 192.168.4.1 via ge-0/1/0.0
> > to 192.168.4.2 via ge-0/1/0.0
> 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> > to 192.168.4.1 via ge-0/1/0.0
> to 192.168.4.2 via ge-0/1/0.0
>
> router C# show policy-options policy-statement load-balancing-policy
> from as-path TEST-AS;
> then {
> load-balance per-packet;
> }
>
> router C# show policy-options as-path TEST-AS
> "64665 .*";
>
> router C# show routing-options forwarding-table
> export load-balancing-policy;
>
> I have even configured "multipath" at BGP but doesn't help still.
>
> I wonder how do I verify whether the per packet load balancing is being
> enabled other than traceroute.
>
> Thanks!
>
> Jonathan.
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>

_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
http://puck.nether.net/mailman/listinfo/juniper-nsp

Load balancing per destination will also accomplish this. Per flow load
balancing is just a more fine-grained way of doing so.

-- steve

-----Original Message-----
From: juniper-nsp-admin@puck.nether.net
[mailto:juniper-nsp-admin@puck.nether.net] On Behalf Of Todd Regonini
Sent: Monday, November 25, 2002 10:45 AM
To: Jonathan Tse; Ben Buxton; juniper-nsp@puck.nether.net
Subject: RE: [j-nsp] per packet/flow load balancing.

"load-balance per-packet" will allow you to load balance traffic across
multiple equal-cost paths. However, this is not done on a per packet
basis. It is done on a per flow basis. Keeping the flows together is a
good thing. You will need to setup multiple streams in your scenario
below to accomplish what you want. Thanks.

T
-----Original Message-----
From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
Sent: Monday, November 25, 2002 8:28 AM
To: Ben Buxton; juniper-nsp@puck.nether.net
Subject: Re: [j-nsp] per packet/flow load balancing.


So that means the "load-balance per-packet" is useless?


----- Original Message -----
From: "Ben Buxton" <b.buxton@planettechnologies.nl>
To: "'Jonathan Tse'" <jonathantse@pacific.net.sg>;
<juniper-nsp@puck.nether.net>
Sent: Tuesday, November 26, 2002 12:22 AM
Subject: RE: [j-nsp] per packet/flow load balancing.


>
> Since the IP2 processor, load balancing per packet is a bit of
> a misnomer. A modern Juniper will load balance per *flow*,
> even though the config says per-packet.
>
> You cannot do true per packet load balancing on a Juniper. Which
> is a good thing at the datarates they're designed for.
>
> So if you're measuring with a single flow, this might explain it.
>
> BB
>
> -----Original Message-----
> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> Sent: maandag 25 november 2002 16:59
> To: juniper-nsp@puck.nether.net
> Subject: [j-nsp] per packet/flow load balancing.
>
>
> Hi,
>
> This has been discussed many times here. Apologize if someone think
this
is
> too simple. But only now I give it a try and without luck.
>
> I have setup this in the test lab:
>
> /-- IBGP -- router A -- EBGP --\
> router C < > router D
> \-- IBGP -- router B -- EBGP --/
>
> I believe there are equal cost paths but the router C doesn't seem to
be
> able the load balance per packet.
>
> Below is a snapshot from router C:
>
> 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> to 192.168.4.1 via ge-0/1/0.0
> > to 192.168.4.2 via ge-0/1/0.0
> 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> > to 192.168.4.1 via ge-0/1/0.0
> to 192.168.4.2 via ge-0/1/0.0
>
> router C# show policy-options policy-statement load-balancing-policy
> from as-path TEST-AS;
> then {
> load-balance per-packet;
> }
>
> router C# show policy-options as-path TEST-AS
> "64665 .*";
>
> router C# show routing-options forwarding-table
> export load-balancing-policy;
>
> I have even configured "multipath" at BGP but doesn't help still.
>
> I wonder how do I verify whether the per packet load balancing is
being
> enabled other than traceroute.
>
> Thanks!
>
> Jonathan.
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>

_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
http://puck.nether.net/mailman/listinfo/juniper-nsp

_______________________________________________
juniper-nsp mailing list juniper-nsp@puck.nether.net
http://puck.nether.net/mailman/listinfo/juniper-nsp

thanks for all the replies!

----- Original Message -----
From: "Todd Regonini" <regonini@juniper.net>
To: "Jonathan Tse" <jonathantse@pacific.net.sg>; "Ben Buxton"
<b.buxton@planettechnologies.nl>; <juniper-nsp@puck.nether.net>
Sent: Tuesday, November 26, 2002 12:45 AM
Subject: RE: [j-nsp] per packet/flow load balancing.


> "load-balance per-packet" will allow you to load balance traffic across
multiple equal-cost paths. However, this is not done on a per packet basis.
It is done on a per flow basis. Keeping the flows together is a good thing.
You will need to setup multiple streams in your scenario below to accomplish
what you want. Thanks.
>
> T
> -----Original Message-----
> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> Sent: Monday, November 25, 2002 8:28 AM
> To: Ben Buxton; juniper-nsp@puck.nether.net
> Subject: Re: [j-nsp] per packet/flow load balancing.
>
>
> So that means the "load-balance per-packet" is useless?
>
>
> ----- Original Message -----
> From: "Ben Buxton" <b.buxton@planettechnologies.nl>
> To: "'Jonathan Tse'" <jonathantse@pacific.net.sg>;
> <juniper-nsp@puck.nether.net>
> Sent: Tuesday, November 26, 2002 12:22 AM
> Subject: RE: [j-nsp] per packet/flow load balancing.
>
>
> >
> > Since the IP2 processor, load balancing per packet is a bit of
> > a misnomer. A modern Juniper will load balance per *flow*,
> > even though the config says per-packet.
> >
> > You cannot do true per packet load balancing on a Juniper. Which
> > is a good thing at the datarates they're designed for.
> >
> > So if you're measuring with a single flow, this might explain it.
> >
> > BB
> >
> > -----Original Message-----
> > From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> > Sent: maandag 25 november 2002 16:59
> > To: juniper-nsp@puck.nether.net
> > Subject: [j-nsp] per packet/flow load balancing.
> >
> >
> > Hi,
> >
> > This has been discussed many times here. Apologize if someone think this
> is
> > too simple. But only now I give it a try and without luck.
> >
> > I have setup this in the test lab:
> >
> > /-- IBGP -- router A -- EBGP --\
> > router C < > router D
> > \-- IBGP -- router B -- EBGP --/
> >
> > I believe there are equal cost paths but the router C doesn't seem to be
> > able the load balance per packet.
> >
> > Below is a snapshot from router C:
> >
> > 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> > 192.168.9.2
> > AS path: 64665 64665 I
> > to 192.168.4.1 via ge-0/1/0.0
> > > to 192.168.4.2 via ge-0/1/0.0
> > 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> > 192.168.9.2
> > AS path: 64665 64665 I
> > > to 192.168.4.1 via ge-0/1/0.0
> > to 192.168.4.2 via ge-0/1/0.0
> >
> > router C# show policy-options policy-statement load-balancing-policy
> > from as-path TEST-AS;
> > then {
> > load-balance per-packet;
> > }
> >
> > router C# show policy-options as-path TEST-AS
> > "64665 .*";
> >
> > router C# show routing-options forwarding-table
> > export load-balancing-policy;
> >
> > I have even configured "multipath" at BGP but doesn't help still.
> >
> > I wonder how do I verify whether the per packet load balancing is being
> > enabled other than traceroute.
> >
> > Thanks!
> >
> > Jonathan.
> >
> > _______________________________________________
> > juniper-nsp mailing list juniper-nsp@puck.nether.net
> > http://puck.nether.net/mailman/listinfo/juniper-nsp
> >
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>

Ben,

--

"Peace cannot be kept by force. It can only be achieved by
understanding." Albert Einstein.


On Mon, 25 Nov 2002, Ben Buxton wrote:

>
> Since the IP2 processor, load balancing per packet is a bit of
> a misnomer. A modern Juniper will load balance per *flow*,

Internet Processor I ASIC forwarded traffic per-packet in a
round-robin fashion among the multiple next-hops.

Internet Processor II ASIC forwards packets based on individual traffic
flows (micro-flows). Packets for each individual flow are kept on a
single interface.

Juniper didn't change the keyboard "per-packet" when they introduced the
IP2 Chip, to keep consistency in the config files


> even though the config says per-packet.
>
> You cannot do true per packet load balancing on a Juniper. Which
> is a good thing at the datarates they're designed for.
>

By default, the hash key use just Layer 3 information (Incoming Interface,
Dest IP, Source IP, IPv4 Prot ID).

As the hash-key control de info that is used int the load-balancing
algorithm you can give even more information to that algorithm if you
enable Layer-4 (Source Port, Dest. Port, TOS Byte).

Hash result is scaled down to the number of available next-hops.
All packets having the same hash will go to the same next-hop interface.

> So if you're measuring with a single flow, this might
explain it. >
> BB
>
> -----Original Message-----
> From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
> Sent: maandag 25 november 2002 16:59
> To: juniper-nsp@puck.nether.net
> Subject: [j-nsp] per packet/flow load balancing.
>
>
> Hi,
>
> This has been discussed many times here. Apologize if someone think this is
> too simple. But only now I give it a try and without luck.
>
> I have setup this in the test lab:
>
> /-- IBGP -- router A -- EBGP --\
> router C < > router D
> \-- IBGP -- router B -- EBGP --/
>
> I believe there are equal cost paths but the router C doesn't seem to be
> able the load balance per packet.
>
> Below is a snapshot from router C:
>
> 192.168.44.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> to 192.168.4.1 via ge-0/1/0.0
> > to 192.168.4.2 via ge-0/1/0.0
> 192.168.48.0/22 *[BGP/170] 19:07:50, MED 90, localpref 350, from
> 192.168.9.2
> AS path: 64665 64665 I
> > to 192.168.4.1 via ge-0/1/0.0
> to 192.168.4.2 via ge-0/1/0.0
>
> router C# show policy-options policy-statement load-balancing-policy
> from as-path TEST-AS;
> then {
> load-balance per-packet;
> }
>
> router C# show policy-options as-path TEST-AS
> "64665 .*";
>
> router C# show routing-options forwarding-table
> export load-balancing-policy;
>
> I have even configured "multipath" at BGP but doesn't help still.
>
> I wonder how do I verify whether the per packet load balancing is being
> enabled other than traceroute.
>
> Thanks!
>
> Jonathan.
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
> _______________________________________________
> juniper-nsp mailing list juniper-nsp@puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>

Hi Jonathan,

On the early Juniper routers with the Internet Processor I
Switch fabrics for the M Series, per packet load balancing was possible.
On the later, more advanced Internet Processor II switch fabric
(more common), per packet load balancing is not possible.
The software knob remains in Junos, but the functionality varies based on
the model of the switch fab in your M Series router.

Thanks,

Bob O'Hara

Systems Engineer

Juniper Networks - 'Every Bit IP'

...........................................
. Email: rohara@juniper.net .
. Web: http://www.juniper.net .
...........................................

-----Original Message-----
From: Jonathan Tse [mailto:jonathantse@pacific.net.sg]
Sent: Monday, November 25, 2002 11:28 AM
To: Ben Buxton; juniper-nsp@puck.nether.net
Subject: Re: [j-nsp] per packet/flow load balancing.


So that means the "load-balance per-packet" is useless?