Switches, Hubs, and Modems
cancel
Showing results for 
Search instead for 
Did you mean: 

Link aggregation issues+questions

SOLVED
Go to solution
Jordi Inglés
Occasional Contributor

Link aggregation issues+questions

Hi all,

 

We have some troubles regarding link aggregation, let me explain whats the setup.

 

What we have:

-HP Procurve 3500yl

-HP Procurve 2910al

-Cisco 3750

 

-6 servers

 

So the setup we have is:

-2 servers connected to the Cisco switch to ports 1 and 2 respectively.The switch has all the ports untagged as vlan 13

-From the cisco switch there are two ethernet cables connected to ports 23 and 24 that connect with the layer3 switch(Procurve 3500yl) to ports 19 and 20.

-On the layer3 there are two machines(port 1 and 2) connected directly, with these ports untagged on vlan 2.

 

This means:

Server1 (192.168.13.1)

Server2 (192.168.13.2)

||

cisco(192.168.13.253)

| |

procurve 3500yl(192.168.13.254/192.168.2.254)

||

Server101(192.168.2.1)

Server101(192.168.2.2)

 

The problem is that doing an iperf from server1 to server101, and from server2 to server102 i get only 500mbps on each server.

 

Relevant parts of the cisco config is:

interface Port-channel1
 switchport access vlan 13
 switchport mode access
!
interface GigabitEthernet3/0/1
 switchport access vlan 13
 switchport mode access
!
interface GigabitEthernet3/0/2
 switchport access vlan 13
 switchport mode access
!

interface GigabitEthernet3/0/23
 switchport access vlan 13
 switchport mode access
 channel-group 1 mode on
!
interface GigabitEthernet3/0/24
 switchport access vlan 13
 switchport mode access
 channel-group 1 mode on
!
interface Vlan13
 ip address 192.168.13.253 255.255.255.0
!

The relevant parts from the hp procurve config is:

trunk 19-20 Trk5 Trunk

trunk 17-18 Trk6 Trunk

vlan 2
   name "V2"
   untagged 1-2
   ip address 192.168.2.254 255.255.255.0
   exit
vlan 12
   name "V12"
   untagged Trk6
   ip address 192.168.12.254 255.255.255.0
   exit
vlan 13
   name "V13"
   untagged Trk5
   ip address 192.168.13.254 255.255.255.0
   exit

spanning-tree Trk5 priority 4
spanning-tree Trk6 priority 4

 

On the other hand, with the following setup:

 

Server991 (192.168.12.1)

Server992 (192.168.12.2)

||

procurve 2910al(192.168.12.253)

| |

procurve 3500yl(192.168.12.254/192.168.2.254)

||

Server101(192.168.2.1)

Server101(192.168.2.2)

I get 2gbps.

 

The relevants part of the configuration on the 2910al is as follows:

trunk 47-48 Trk1 Trunk

vlan 12
   name "v12"
   untagged 1-46,Trk1
   ip address 192.168.12.253 255.255.255.0
   exit
spanning-tree Trk1 priority 4
primary-vlan 12

So what i'm doing wrong on the cisco?

 

Thank you in advance, i've been fighting with this for months with no solution...

4 REPLIES
Oleg Sukharev
Valued Contributor
Solution

Re: Link aggregation issues+questions

Hi Jodi,

Have you tried
port-channel load-balance src-dst-ip
on Cisco?

By the way ProCurve and Cisco support LACP trunks =)

hth,
oleg
Jordi Inglés
Occasional Contributor

Re: Link aggregation issues+questions

Hi Oleg,

 

Well, the default load balancing algorithm is "Source MAC address", and if the load algorithm checks the MAC of the machines and not the ones on the switches, there would be 4 MAC(2 source, 2 remote) addresses. Nevertheless i tried it anyway, same result.

 

And with LACP the result is the same as well.

 

Thanks!

Oleg Sukharev
Valued Contributor

Re: Link aggregation issues+questions

Jordi,

 

Please try running command "show interface display" on 3500yl while using iperf  to find who is not doing balancing. The command shows dynamic table for all ports.

 

It can be that ProCurve puts the two streams to the same link based on last bits of source and destination IPs. So another tests can be: 1) change IP of one of the servers (say increase by 1), 2) try command "trunk-load-balance L4-based" on 3500yl.

 

hth,

oleg

Jordi Inglés
Occasional Contributor

Re: Link aggregation issues+questions

Hi Oleg,

 

although none of the commands you've told me are available on the 3500yl, i've found the issue. One of the clients had a wrong IP, after changing it to the correct one, it worked fine.

 

Thanks!