Discussion:
[j-nsp] LACP hashing algorithm
junos fordummies via juniper-nsp
2018-08-08 23:32:03 UTC
Permalink
Hi all,

This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2

Just wondering if anyone has seen anything similar ?

Many thanks

JfD.
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Richard McGovern
2018-08-09 10:57:56 UTC
Permalink
Yes, have seen this, but need to check how we resolved this. Will get back to you.

On 8/8/18, 7:32 PM, "junos fordummies" <***@googlemail.com> wrote:

Hi all,

This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2

Just wondering if anyone has seen anything similar ?

Many thanks

JfD.



_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Luis Balbinot
2018-08-09 11:17:56 UTC
Permalink
How many flows are there in total? Is there a test appliance involved? We
had many issues with those in the past during service delivery tests.

Also I assume you are using MPCs and not DPCs and also that you are talking
about IP traffic. Please correct me if not.

Luis

On Wed, 8 Aug 2018 at 20:32 junos fordummies via juniper-nsp <
Post by junos fordummies via juniper-nsp
Hi all,
This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2
Just wondering if anyone has seen anything similar ?
Many thanks
JfD.
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Richard McGovern
2018-08-09 16:39:30 UTC
Permalink
I am assuming the traffic has MPLS labels, yes??

On 8/9/18, 7:17 AM, "Luis Balbinot" <***@luisbalbinot.com> wrote:

How many flows are there in total? Is there a test appliance involved? We
had many issues with those in the past during service delivery tests.

Also I assume you are using MPCs and not DPCs and also that you are talking
about IP traffic. Please correct me if not.

Luis

On Wed, 8 Aug 2018 at 20:32 junos fordummies via juniper-nsp <
Post by junos fordummies via juniper-nsp
Hi all,
This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2
Just wondering if anyone has seen anything similar ?
Many thanks
JfD.
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
junos fordummies via juniper-nsp
2018-08-10 18:42:56 UTC
Permalink
Guys thanks for all the replies, when I took an sflow sample, I oculd see
at least 100+ flows with varying src/dst tuple and flow sizes, so cant
understand why flows are stuck on one link. I can understand if tuples were
hashing to one link, but these are varies sources/destinations... makes no
sense.

I will have to raise a JTAC as I cant for the life of riley figure out
whats happening. I have double checked the config, will also look at what
Niall posted.

I will come back to you once JTAC have a look.

Thanks

JFD
Post by Luis Balbinot
How many flows are there in total? Is there a test appliance involved? We
had many issues with those in the past during service delivery tests.
Also I assume you are using MPCs and not DPCs and also that you are
talking about IP traffic. Please correct me if not.
Luis
On Wed, 8 Aug 2018 at 20:32 junos fordummies via juniper-nsp <
Post by junos fordummies via juniper-nsp
Hi all,
This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2
Just wondering if anyone has seen anything similar ?
Many thanks
JfD.
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
junos fordummies via juniper-nsp
2018-08-10 18:57:01 UTC
Permalink
ok update.

We restarted the MPC4E on our 960 and voila, all 4 links being used- LOL
!!! The MPC4E that we restarted also had another AE bundle exhibiting the
same issue across other ports. Anyways, working now, cant even begin to
explain what the problem might have been.

Many thanks

JfD

On 10 August 2018 at 19:42, junos fordummies <
Post by junos fordummies via juniper-nsp
Guys thanks for all the replies, when I took an sflow sample, I oculd see
at least 100+ flows with varying src/dst tuple and flow sizes, so cant
understand why flows are stuck on one link. I can understand if tuples were
hashing to one link, but these are varies sources/destinations... makes no
sense.
I will have to raise a JTAC as I cant for the life of riley figure out
whats happening. I have double checked the config, will also look at what
Niall posted.
I will come back to you once JTAC have a look.
Thanks
JFD
Post by Luis Balbinot
How many flows are there in total? Is there a test appliance involved? We
had many issues with those in the past during service delivery tests.
Also I assume you are using MPCs and not DPCs and also that you are
talking about IP traffic. Please correct me if not.
Luis
On Wed, 8 Aug 2018 at 20:32 junos fordummies via juniper-nsp <
Post by junos fordummies via juniper-nsp
Hi all,
This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2
Just wondering if anyone has seen anything similar ?
Many thanks
JfD.
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Catalin Dominte
2018-08-12 18:06:27 UTC
Permalink
Hello,

I saw a similar issue, and the JTAC traced it to a software issue whereby a LACP member would be incorrectly programmed on the ASIC for random VLAN tags. Disable / Re-enable the interface did not fix it, but removing the port from LACP and adding it back fixed the ASIC programming issue.

This seems pretty similar to the issue I saw.


Catalin Dominte
Senior Network Consultant

Nocsult Ltd | 2 Cambridge House | Gogmore Lane | Chertsey | KT16 9AP | Phone: +44 (0)1628 302 007
VAT registration number: GB 180957674 | Company registration number: 08886349
P Please consider the environment - Do you really need to print this email?

THIS COMMUNICATION MAY CONTAIN CONFIDENTIAL AND/OR OTHERWISE PROPRIETARY MATERIAL and is thus for use only by the intended recipient. If you received this in error, please contact the sender and delete the email and its attachments from all computers.
Post by junos fordummies via juniper-nsp
ok update.
We restarted the MPC4E on our 960 and voila, all 4 links being used- LOL
!!! The MPC4E that we restarted also had another AE bundle exhibiting the
same issue across other ports. Anyways, working now, cant even begin to
explain what the problem might have been.
Many thanks
JfD
On 10 August 2018 at 19:42, junos fordummies <
Post by junos fordummies via juniper-nsp
Guys thanks for all the replies, when I took an sflow sample, I oculd see
at least 100+ flows with varying src/dst tuple and flow sizes, so cant
understand why flows are stuck on one link. I can understand if tuples were
hashing to one link, but these are varies sources/destinations... makes no
sense.
I will have to raise a JTAC as I cant for the life of riley figure out
whats happening. I have double checked the config, will also look at what
Niall posted.
I will come back to you once JTAC have a look.
Thanks
JFD
Post by Luis Balbinot
How many flows are there in total? Is there a test appliance involved? We
had many issues with those in the past during service delivery tests.
Also I assume you are using MPCs and not DPCs and also that you are
talking about IP traffic. Please correct me if not.
Luis
On Wed, 8 Aug 2018 at 20:32 junos fordummies via juniper-nsp <
Post by junos fordummies via juniper-nsp
Hi all,
This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2
Just wondering if anyone has seen anything similar ?
Many thanks
JfD.
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp

Pavel Lunin
2018-08-10 19:21:18 UTC
Permalink
To troubleshoot this kind of condition, you need to understand 1) the
complete structure of the headers (is there any tunneling, MPLS,
pseudowires etc) 2) what kind of forwarding decision your MX performs for
those packets: IP LPM only, Ethernet switching, IP + Ethernet (irb-based
L3), MPLS, MPLS+IP ect.

If you have Ethernet bridging towards this interface, it might be a result
of unknown unicast or some other sort of flooding (e. g. multicast dst MAC).

Also cheek the [edit forwarding-options enhanced-hash-key] configuration.
People often tend to put nonsense there.

If it resembles a bug, there is a uKernel command to check the "really"
(from the MPC's point of view) applied hash-key config. Being far from my
PC I don't remember the exact syntax, you can find it at least in David
Roy's brochure "an expert packet walkthrough on the mx series 3d" or
someone here might remind us.

--
Pavel

пт, 10 авг. 2018 г., 20:43 junos fordummies via juniper-nsp <
Post by junos fordummies via juniper-nsp
Guys thanks for all the replies, when I took an sflow sample, I oculd see
at least 100+ flows with varying src/dst tuple and flow sizes, so cant
understand why flows are stuck on one link. I can understand if tuples were
hashing to one link, but these are varies sources/destinations... makes no
sense.
I will have to raise a JTAC as I cant for the life of riley figure out
whats happening. I have double checked the config, will also look at what
Niall posted.
I will come back to you once JTAC have a look.
Thanks
JFD
Post by Luis Balbinot
How many flows are there in total? Is there a test appliance involved? We
had many issues with those in the past during service delivery tests.
Also I assume you are using MPCs and not DPCs and also that you are
talking about IP traffic. Please correct me if not.
Luis
On Wed, 8 Aug 2018 at 20:32 junos fordummies via juniper-nsp <
Post by junos fordummies via juniper-nsp
Hi all,
This will sound like a very weird question, but has anyone seen a
scenario
Post by Luis Balbinot
Post by junos fordummies via juniper-nsp
whereby an MX960 with 4 x 10G links always hashes (uses) a single link
out
Post by Luis Balbinot
Post by junos fordummies via juniper-nsp
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2
Just wondering if anyone has seen anything similar ?
Many thanks
JfD.
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/jun
Eduardo Schoedler
2018-08-09 16:16:03 UTC
Permalink
Have you tried load-balance adaptive?

set interfaces aeX aggregated-ether-options load-balance adaptive tolerance 10


Regards,

2018-08-08 20:32 GMT-03:00 junos fordummies via juniper-nsp
Post by junos fordummies via juniper-nsp
Hi all,
This will sound like a very weird question, but has anyone seen a scenario
whereby an MX960 with 4 x 10G links always hashes (uses) a single link out
of the 40G bundle ? We have restarted the device, traffic flows in one
direction only use a single link, the reverse path is all 4 links in the
bundle. Both ends of the AE is MX960 running 16.1r2
Just wondering if anyone has seen anything similar ?
Many thanks
JfD.
_______________________________________________
https://puck.nether.net/mailman/listinfo/juniper-nsp
--
Eduardo Schoedler
_______________________________________________
juniper-nsp mailing list juniper-***@puck.nether.net
https://puck.nether.net/mailman/listinfo/juniper-nsp
Loading...