Comware

Hi, 

A customer has 2 X WAN circuits linking two Hospitals

The WAN circuits terminate on their HPE Comware L3-CORE in either location

At present, we have configured a /29 IP transit networks on either WAN circuit and use OSPF ECMP to load-balance traffic across the two WAN circuits

Switchport config pasted below

Hospital-A-CORE-SW

<CORE-5945-STACK>display current-configuration interface Twenty-FiveGigE 1/0/43
#
interface Twenty-FiveGigE1/0/43
 port link-mode route
 description ## 2GB WAN-1 circuit CORE-5710-STACK 1/0/47 ##
 bandwidth 2000000
 speed 10000 
 mtu 9000
 jumboframe enable 9000
 ip address 172.27.4.17 255.255.255.248
 ospf cost 6
 ospf network-type p2p
 sflow flow collector 1
 sflow sampling-rate 2000
#
return
<CORE-5945-STACK>display current-configuration interface Twenty-FiveGigE 3/0/43
#
interface Twenty-FiveGigE3/0/43
 port link-mode route
 description ## 2GB WAN-2 circuit CORE-5710-STACK 2/0/47 ##
 bandwidth 2000000
 speed 10000 
 mtu 9000
 jumboframe enable 9000
 ip address 172.27.4.25 255.255.255.248
 ospf cost 6
 ospf network-type p2p
 sflow flow collector 1
 sflow sampling-rate 2000
#

Hospital-B-CORE-SW

<CORE-5710-STACK>display current-configuration int Ten-GigabitEthernet 1/0/47
#
interface Ten-GigabitEthernet1/0/47
 port link-mode route
 description ## 2GB WAN-1 CORE-5945-STACK WGE1/0/43 ##
 bandwidth 2000000
 mtu 9000
 jumboframe enable 9000
 ip address 172.27.4.22 255.255.255.248
 ospf cost 6
 ospf network-type p2p
 sflow flow collector 1
 sflow sampling-rate 2000
#
return
<CORE-5710-STACK>display current-configuration int Ten-GigabitEthernet 2/0/47
#
interface Ten-GigabitEthernet2/0/47
 port link-mode route
 description ## 2GB WAN-2 CORE-5945-STACK WGE3/0/43 ##
 bandwidth 2000000
 mtu 9000
 jumboframe enable 9000
 ip address 172.27.4.30 255.255.255.248
 ospf cost 6
 ospf network-type p2p
 sflow flow collector 1
 sflow sampling-rate 2000
#

This works fine so far (2 years)

However, the circuits are getting busier and for specific source-ip and destination-ip sessions (e.g. VxLAN tunnel between their VTEPs) , the OSPF ECMP algorithm favours using one of the WAN circuits primarily. This is because ECMP does not mean load-balancing. Based on hash algorithm it will choose the next hop. If the source and destination is always same (e.g. a VxLAN tunnel) the nexthop chosen across two WAN circuits will always be same

We're trying to address this ECMP behaviour because the bulk of the inter-site traffic is storage replication within that VxLAN tunnel

My idea was to bundle the two WAN circuits into a single "Route-Aggregation Group" , instead of having two separate IP transit links

We could re-configure the 4 switchports on either end of the two 2GB WAN circuits to be part of the same "Route-Aggregation Group"  and then take advantage of improved load-balancing

My question is about the load-balancing algorithm associated with Comware "Route-Aggregation Groups" – is it superior to the OSPF ECMP  algorithm ?

In other words, would the VxLAN tunnel traffic between the VTEP endpoints 172.27.4.202 ßà 172.26.254.202 be more evenly shared between the two WAN circuits, using a Route-aggregation Group instead of OSPF ECMP ?

Thanks for any advice

Hi

I see that :

172.27.4.17/29 and 172.27.4.22/29 are on the same subnet

172.27.4.25/29 and 172.27.4.30/29 are on the same other subnet

What is the WAN links type ? Direct fiber between the two hospitals ? If so, why not agregate the links at L2 ?

If not (there is an L3 operator in-betwwen), I don't understand how you can have the same subnets on both sides of each link...

Hi Frederic, 

172.27.4.17/29 and 172.27.4.22/29 are opposite ends of the first WAN circuit (WAN Circuit-A)

172.27.4.25/29 and 172.27.4.30/29 are opposite ends of the second WAN circuit (WAN Circuit-B)

To answer your questions

What is the WAN links type ? Direct fiber between the two hospitals ? If so, why not agregate the links at L2 ?

The WAN circuits are both Layer2 circuits. We can do whatever we like with them, the service provider doesn't care. 

We've chosen to configure IP addressing on either end and the switchports are configured in "route-mode" , because we do not want any Layer2 spanning-tree interaction between the two Hospital LANs

Yes, we are planning to aggregate both Links at L3 (using a route-aggregation Group) , rather than aggregate them at L2 (into a bridge-aggregation group)

Again here , the reason for this is ecause we do not want any Layer2 spanning-tree interaction between the two Hospital LANs

Capiche ?