04-18-2016 03:30 AM
following my previous question about ICL interconnect between 2 DCX 8510-8:
when it comes to Ultra Scale ICL interconnect between 2 Directors, Brocade recommends that there are minimum of 2 ports on each core blades used
To connect multiple Brocade DCX 8510 chassis via UltraScale ICLs, a minimum of four ICL ports (two on each
core blade) must be connected between each chassis pair
The dual connections on each core blade must reside within the same ICL trunk boundary on the core blades. If more than four ICL
connections are required between a pair of Brocade DCX 8510 chassis, additional ICL connections should be
added in pairs (one on each core blade).
and one more recommendation I remember, but can not find the quote is that if only two directors are connected, then no more than 4 ports on each core blade should be used.
Anyway, the customer has now 5 ICLs per Core blade which is not best practice. I wonder what disadvantage may it have and is there a way to detach this 5th ICL without distrupting traffic? Or there is no need for 5th ICL removal?
04-19-2016 12:44 AM
by the way, one more fact.
5th ICL is placed in another Trunk group and therefore forms 4 new Trunks with 1ISL member only in every Trunk.
So, at the end there are 16 Trunks between two DCX, 8 Trunks with 4 ISL members and 8 Trunks with only 1 ISL member.
Interesting how frames will be routed across these 16 Trunks
04-27-2016 01:50 PM
As mentioned in your post the best practice for interconnecting 2 DCX 8510-8 is to have minimum 4 ICL and spread them among CR16-8 blades. The advantage of such approach is that we have redundancy - 2 core routing blades involved from each chassis. Loosing one blade still we have another working. Thus we sacrifice the trunk bandwith for the redundant configuration.
The answer for your second question is yes, you could use portdecom command to disable an E_Port without frame loss. Please refer to FOS_CmdRef for more information on this.
Hope this helps.
04-27-2016 02:00 PM
Brocade uses DLS and DPS extensions to FSPF for frame routing. With DPS and DLS are ON which is the default frames of the same Sequence will go to the same ISL or Trunk. The choice of the next ISL or Trunk for Sequence is based on S_ID (source address ), D_ID (destination address) and OX_ID (Originator Exchange ID). Thus different scsi commands even from same Initiator or Target are distributed among links. Moreover DLS will consider the bandwith so trunk will be choosen more often. And the last - frames of the sequence will be distributed among the physical ISLs within the trunk.