MLAG algorithm when connected to Juniper Switches.

  • 0
  • 1
  • Question
  • Updated 3 years ago
  • Answered
I have a customer who has a core of Extreme , and Juniper edge .

We seem to have an issue with Mlag.
The config looks OK , only thing different to my configuration is that the ISC is in its own VR, I do not see an issue with that ..

The problem that is being reported , is that the edge switches ( juniper when configured for AE0, the transfer of data is extremely slow.

Remove one of the mlag links and it all speeds up to an acceptable level.

We are not running LACP.. 

Juniper Edge  config.

set interfaces ae0 description xe-uplinks-LAG set interfaces ae0 aggregated-ether-options minimum-links 1
set interfaces ae0 aggregated-ether-options link-speed 10g
set interfaces ae0 unit 0 family ethernet-switching port-mode trunk 

set interfaces xe-0/1/0 ether-options 802.3ad ae0 
set interfaces xe-0/1/1 ether-options 802.3ad ae0


Extreme config :

One side only ..

create vr "VR-MLAG-ISC" 
create vlan "MLAG-ISC" vr VR-MLAG-ISC
configure vlan MLAG-ISC tag 6
configure vlan "MLAG-ISC"  qosprofile QP5
configure vlan MLAG-ISC add ports 1:39 tagged  
configure vlan MLAG-ISC ipaddress 10.10.10.1 255.255.255.252
disable iproute ipv4 compression vr VR-MLAG-ISC
disable iproute ipv6 compression vr VR-MLAG-ISC
create mlag peer "CL-MLAG" 
configure mlag peer "CL-MLAG" ipaddress 10.10.10.2 vr VR-MLAG-ISC  enable mlag port 1:43 peer "CL-MLAG" id 1 
enable mlag port 1:44 peer "CL-MLAG" id 2 
enable mlag port 1:45 peer "CL-MLAG" id 3 
enable mlag port 1:46 peer "CL-MLAG" id 4 
enable mlag port 1:47 peer "CL-MLAG" id 5 
enable mlag port 1:48 peer "CL-MLAG" id 6 

 Ok this is more than I would configure normally for MLAG..(QOS IProute Compression )..

My main question is :

What hashing algorithm does Mlag use , its construction and can it be modified , 

I expect the issue to be the difference between  Juniper and Extreme Algorithms ... well I hope it is , if this cannot be resolved , the customer will be returning to EAPS.

Many Thanks

Rod
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb

Posted 3 years ago

  • 0
  • 1
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,594 Points 10k badge 2x thumb
Hi Rod,

Can you please explain why the ISC was put into its own VR?
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
 LOL This was not installed by me, though I have to support it .. and I cant ask the original engineer as they are long gone.

Normally with Mlag I dont seem to get these problems when running extreme everywhere.

The answer is I dont know why they created a seperate VR  normally I would just put it in  the VR-default ( which at the end of the day is just another VR right !!! ? )

The issue we are having is that when mlag is in and working ( or not )  the performance on the links to and from the juniper switches is unworkable ..remove one of the link , hay presto all works OK.
Photo of Henrique

Henrique, Employee

  • 10,302 Points 10k badge 2x thumb
Hi Rob, I don't see the sharing enabled (link-agregation) on EXOS switch.

Could you please share this configuration?

You must enable sharing in EXOS port connecting to Juniper even if using just one link.
Photo of Henrique

Henrique, Employee

  • 10,302 Points 10k badge 2x thumb
Just a correction. This is needed only when using LACP, not for Static LAG.
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
Hi We are using static lag on the juniper and no LACP.
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,594 Points 10k badge 2x thumb
Hi Rod,

At first glance I believe that ISC being put into a different VR is in fact causing the issue you are seeing. It is probably causing an issue with the blocking MLAG is supposed to perform. Would it be possible to change it? I don't see any advantage to having the ISC in a different VR.
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
OK , if we have multiple VR 's , and of course the edge switch does not support vr's AKA X440's , are you implying that we should have an ISC per VR ?
Photo of Frank

Frank

  • 3,776 Points 3k badge 2x thumb
We have multiple VRs and only one ISC. On another 8800 pair we have the ISC in its own VR (because the rest of the vlans that were there moved to another BD pair)
I'd look at the  "shared" part that Henrique mentioned.
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,594 Points 10k badge 2x thumb
The shared port is only needed if you are using LACP.
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
We have 11 VR's on this PAir of X670 switches ..
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
So basically it does not matter what vr we use for the ISC .. normally I would just have the ISC in VR-Default.. it works as I have done this many times ..
(Edited)
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,594 Points 10k badge 2x thumb
Hi Rod,

When you have this second port active how does the TOP output look? Is there a particular process running high on either of the MLAG switches?
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
Im hoping to get out there tomorrow ... what am I looking for MXBCM ?
Photo of Henrique

Henrique, Employee

  • 10,302 Points 10k badge 2x thumb
Rod, when configuring 802.3ad on Juniper side, is there any algorithm option like L2, L3, L3_L4 similar to those in EXOS?

Just want to share that I saw a similar behavior (slow data transfer) last year in another vendor connecting to an MLAG pair and it was fixed using L3 sharing algorithm on both link ends.

In that case both side of the links were using Static LAG even using 1 port (I know this is not needed) and we changed the algorithm on both ends to a higher level (L3/L3_L4).

Hope this give some clue.
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
Hi

Ok so the suggestion for the uplink 1:43 is to set it to sharing with only one member and then change the hash to l3 .. sort of what you do mlag when using LACP
Photo of Henrique

Henrique, Employee

  • 10,302 Points 10k badge 2x thumb
Hi Rod, I'm not sure if that will fix the issue. However, that change has fixed a similar problem last year.

I know this is mandatory when configuring LACP only, but it's a good option to try.
Photo of Grosjean, Stephane

Grosjean, Stephane, Employee

  • 13,284 Points 10k badge 2x thumb
I'm a bit late on that topic, but are you sure to have matching MLAG Id on both MLAG peers?
Photo of Rod Robertson

Rod Robertson

  • 2,344 Points 2k badge 2x thumb
I have now been to site and the actual issue seen is different to that originally described by myself.

This is basically a traditional mlag using 3 sets of X670 ( 2 x stack  ) 8 x 10 Gig connection between the mlag peers and 2 X 10 Gig conenctions from each peer to the stacked X670 edge.

This is running 15.5.5.2 code due to a bug giving igmpv3 issues , we have already tried upgrading to 15.6.3.1-patch1-9 , though this just returned us to the original reason for needing the bug fix 15.5.5.2

Basically when they are having issues , ie cannot use the network ( though not everyone is effected they remove on of the peer uplinks and it all works 



Currently they have disconnected 2 links as in diagram above
Photo of Patrick Voss

Patrick Voss, Alum

  • 11,594 Points 10k badge 2x thumb
Can you please send a show sharing from the top stack in the diagram?
Photo of Henrique

Henrique, Employee

  • 10,302 Points 10k badge 2x thumb
That issue might be related to wrong LAG/MLAG configuration.

Please send the following outputs:

show sharing (already requested by Patrick) for all 3 stacks
show edp ports all for all 3 stacks so we can check the # of vlans added on each LAG/MLAG uplink
show configuration vsm for both MLAG peers