Distributed Trunking with 4 6600-24XG Switches

[deflowjo]

Hello,

i want to set this up with 4 switches

2 pairs of 660024XG 

switch 2&3 = pair A, isc trunk1 on port 23,24

switch 4&5 = pair B, isc trunk1 on port 23,24

both pairs setup with a peer keepalive and switch interconnect

now; if i put the first pair and configure a dt trunk from pair A(each port 22) to switch 4 on port 21,22 

next i would configure a dt trunk to switch 5 from pair A(each port 21) to switch 4 on port 21,22

so i have a trunk between switch2 & 3 = pair A, from there i have a dt trunk to switch 4 (call it trunk A4), and a dt trunk to switch 5 (call it trunk A5)

switch 4&5 also form a pair and have their own trunk between them

still there is a loop, because of the trunk between pair B, how do i set this up?the white paper only talks about the commands, an example shows only trunk between 3 switches.

i want to be able to set up dt trunk from pair A to a final edge switch/server, and from pair B to another final switch/server, like in the picture

please help?

on the hp whitepaper picture it looks nice, but dt trunk only goes from a pair of switches to a single switch,  is it possible to put all the ports in the same dt-trunk1, i mean, connect port 21,22 from each switch on pairA, configure as dt-trunk 1, and put the ports on pair B also in dt-trunk 1, creating 1 big dt-trunk1?

unfortunately i could only test dt with 2 switches, to an edge server, now setup is ready and in production with mstp enabled,  i added 2 more swithces, and i am confused,i would use dt-lacp

if anyone could give an insight on this please?

[deflowjo]

i've managed to set this up, with 4 switches

i used 2 ports on each switch for interconnect trunk (that makes 2 pairs of interconnected switches)

i used 2 ports on each switch and put them in dt-lacp (i have put dt-lacp on the source and destination)

then i connecteded thos dt-lacp ports to each other (connected the 4 switches all together)

i have noticed that any, non dt-lacp connection between switch would cause a huge broadcast storm, it is better to just unplug or disable all the other ports

config is like this on all 4 switches

trk1, port 23 and 24 for lacp interconnect (2 cables between a pair of switches for the switch interconnect)

trk2 dt-lacp port 21 and 23, i configured dt-lacp on all 4 switches

heartbeat port 20

is this correct?

result of sh lacp distributed on all 4 switches is

Local Port Status:

            LACP         Trunk   Port                          LACP     Admin  Oper  

Port   Enabled  Group   Status     Partner Status   Key       Key  

----  -------      -------  -------     ------- ------- ------ ------  

23       Active        Trk2       Up           Yes         Success 0            291  

24       Active        Trk2       Up           Yes         Success 0            291

Remote Port Status:

            LACP      Trunk     Port                LACP      Oper   

Port   Enabled   Group     Status    Partner   Status    Key   

----   -------   -------   -------   -------   -------   ------   

23     Active         Trk2      Up         Yes                      Success   0   

24     Active         Trk2      Up         Yes        Success   0

so by configuring the dt-lacp on all the ports that connect to switches it makes 1 big Trk2

i was confused by the fact that you would set up dt-lacp on interconnected switch, and normal lacp on the destination, but the destination is also an interconnected switch, so i used dt-lacp on source ports and destination ports

is this the correct way?

by the way, i use this to make 4 6600 switches as 1 'stack', but is this the right way to do for a backbone to connect the storage to the servers?

top two switches are connected to storage, bottom two switches are connected to flex-10, and dl 580

next thing to do is use the dt-lacp to make trunks to the storage servers on the top switches, and also use dt-lacp to enable active/active shared uplink sets.

the dl580 have separate physical ports, i will not put them in trunk, but rely on the mpio only