OSPF load-sharing is not working

[hsien-ping tang]

Dear Experts,

I have a question about load-sharing in OSPF 。

There are two equal-cost routes in the routing tale, in my idea, the forwarding behavior should be load-sharing. however, the traffic is forwarding on one uplink.

Below is "show" info:

RP/0/RSP0/CPU0#sh run router ospf 2600
Thu May 17 15:54:13.318 GMT
router ospf 2600
nsr
log adjacency changes
router-id 10.130.127.3
nsf ietf
redistribute connected route-policy CONN
area 0.0.137.11
interface Bundle-Ether2601
network point-to-point
fast-reroute per-link
!
interface Bundle-Ether2602
network point-to-point
fast-reroute per-link
!
!
!

RP/0/RSP0/CPU0:#sh route 10.156.24.128
Thu May 17 15:54:22.287 GMT

Routing entry for 10.156.24.128/32
Known via "ospf 2600", distance 110, metric 3
Tag 3489678349, type extern 1
Installed May 17 11:02:27.300 for 04:51:55
Routing Descriptor Blocks
10.158.26.33, from 10.156.1.11, via Bundle-Ether2601
Route metric is 3
10.158.26.37, from 10.156.1.12, via Bundle-Ether2602
Route metric is 3
No advertising protos.

RP/0/RSP0/CPU0:#show cef 10.156.24.128
Thu May 17 15:54:34.956 GMT
10.156.24.128/32, version 302721, internal 0x1000001 0x0 (ptr 0x726f412c) [1], 0x0 (0x71ce5a78), 0x0 (0x0)
Updated May 17 11:02:27.301
Prefix Len 32, traffic index 0, precedence n/a, priority 1
via 10.158.26.33/32, Bundle-Ether2601, 4 dependencies, weight 0, class 0 [flags 0x0]
path-idx 0 NHID 0x0 [0x716a4654 0x0]
next hop 10.158.26.33/32
local adjacency
via 10.158.26.37/32, Bundle-Ether2602, 3 dependencies, weight 0, class 0, platform-non_preferred [flags 0x0]
path-idx 1 NHID 0x0 [0x716a55c0 0x0]
next hop 10.158.26.37/32
local adjacency

RP/0/RSP0/CPU0:#show cef 10.156.24.128 hardware egress
Thu May 17 15:54:42.204 GMT
10.156.24.128/32, version 302721, internal 0x1000001 0x0 (ptr 0x726f412c) [1], 0x0 (0x71ce5a78), 0x0 (0x0)
Updated May 17 11:02:27.302
Prefix Len 32, traffic index 0, precedence n/a, priority 1
via 10.158.26.33/32, Bundle-Ether2601, 4 dependencies, weight 0, class 0 [flags 0x0]
path-idx 0 NHID 0x0 [0x716a4654 0x0]
next hop 10.158.26.33/32
local adjacency
via 10.158.26.37/32, Bundle-Ether2602, 3 dependencies, weight 0, class 0, platform-non_preferred [flags 0x0]
path-idx 1 NHID 0x0 [0x716a55c0 0x0]
next hop 10.158.26.37/32
local adjacency

RP/0/RSP0/CPU0:#show cef 10.156.24.128 detail
Thu May 17 15:54:50.822 GMT
10.156.24.128/32, version 302721, internal 0x1000001 0x0 (ptr 0x726f412c) [1], 0x0 (0x71ce5a78), 0x0 (0x0)
Updated May 17 11:02:27.302
Prefix Len 32, traffic index 0, precedence n/a, priority 1
gateway array (0x71aeeb58) reference count 120, flags 0x0, source rib (7), 0 backups
[121 type 3 flags 0x8401 (0x71bb48d0) ext 0x0 (0x0)]
LW-LDI[type=3, refc=1, ptr=0x71ce5a78, sh-ldi=0x71bb48d0]
gateway array update type-time 1 Dec 29 01:00:57.071
LDI Update time Dec 29 01:00:57.071
LW-LDI-TS May 17 11:02:27.302
via 10.158.26.33/32, Bundle-Ether2601, 4 dependencies, weight 0, class 0 [flags 0x0]
path-idx 0 NHID 0x0 [0x716a4654 0x0]
next hop 10.158.26.33/32
local adjacency
via 10.158.26.37/32, Bundle-Ether2602, 3 dependencies, weight 0, class 0, platform-non_preferred [flags 0x0]
path-idx 1 NHID 0x0 [0x716a55c0 0x0]
next hop 10.158.26.37/32
local adjacency

Load distribution: 0 (refcount 121)

Hash OK Interface Address
0 Y Bundle-Ether2601 10.158.26.33

I would like to ask experts that

1. what does "platform-non_preferred" mean in show cef?

2. and how come it will be appeared?

3. Is this symptom caused by configuring IP FRR?

Thank you,

Tom

[xr-escalation]

Try configuring 'maximum-paths' under OSPF. That should enable ECMP.

/Aleksandar

[hsien-ping tang]

Hi Aleksander

Isnt it enabled by default?

[xr-escalation]

actually, yes. :) Are you taking these commands from a cluster by any chance? Which platform is this?

[hsien-ping tang]

Hi

I am using ASR9010(nVEdge Cluster) with IOSXR 5.3.3.

Because this Router has service traffic on it, I don,t have the authorization to change config. except show command.

thank you

[xr-escalation · Accepted solution]

In cluster we always prefer interfaces on the local rack. That's what platform-non_preferred is used for. Are members of Bundle-Ether2601 and Bundle-Ether2602 distributed across both racks?

[hsien-ping tang]

Hi

Thanks for the reply,

member info of Bundle-Ethernet 2601 & 2602 shown as below,

RP/0/RSP0/CPU0:#show bundle bundle-ether 2601
Fri May 18 10:18:30.357 GMT

Bundle-Ether2601
Status: Up
Local links <active/standby/configured>: 1 / 0 / 1
Local bandwidth <effective/available>: 1000000 (1000000) kbps
MAC address (source): a80c.0d32.631b (Configured)
Inter-chassis link: No
Minimum active links / bandwidth: 1 / 1 kbps
Maximum active links: 64
Wait while timer: 2000 ms
Load balancing: Default
LACP: Operational
Flap suppression timer: Off
Cisco extensions: Disabled
Non-revertive: Disabled
mLACP: Not configured
IPv4 BFD: Not configured

Port Device State Port ID B/W, kbps
-------------------- --------------- ----------- -------------- ----------
Gi0/7/1/17 Local Active 0x8000, 0x0019 1000000
Link is Active
RP/0/RSP0/CPU0:#show bundle bundle-ether 2602
Fri May 18 10:18:36.034 GMT

Bundle-Ether2602
Status: Up
Local links <active/standby/configured>: 1 / 0 / 1
Local bandwidth <effective/available>: 1000000 (1000000) kbps
MAC address (source): a80c.0d32.631b (Configured)
Inter-chassis link: No
Minimum active links / bandwidth: 1 / 1 kbps
Maximum active links: 64
Wait while timer: 2000 ms
Load balancing: Default
LACP: Operational
Flap suppression timer: Off
Cisco extensions: Disabled
Non-revertive: Disabled
mLACP: Not configured
IPv4 BFD: Not configured

Port Device State Port ID B/W, kbps
-------------------- --------------- ----------- -------------- ----------
Gi1/7/1/17 Local Active 0x8000, 0x0023 1000000
Link is Active

If it prefers interfaces on the local rack in cluster, in our design, it should even be load-sharing, isn't it?

[xr-escalation]

No, that's the point of preferring local interfaces. We don't want to load-balance if that includes sending packets over IRL.

[dllivipuma]

Is there any way to load-balance including the IRL?

[smilstea]

Do you mean load-share when traffic has to traverse the IRLs from one rack to the other in cluster?

This is handled automatically, and all IRLs are treated like they are in 1 bundle, so we use bundle hashing to achieve load balancing. You cannot put the IRLs in a bundle themselves though.

Sam