-
Notifications
You must be signed in to change notification settings - Fork 22
Description
In current rally_ovs
implementation, the logical ports are evenly distributed onto all the emulated chassis.
The pseducode of existing implementation in rally_ovs is as follows
num-lports-per-network = // some number of rally task file
for network in logical-networks
for i=0 to num-lports-per-network
// bind lport[i] to chassis [i]
Therefore, for each virtual network (e.g., having 400 lports), the
current binding strategy binds one lports to one chassis. If the
number of lport is smaller than the # chassis, the allocation tends to
be unbalanced. That's the main reason I choose 400 ports per network.
The binding strategy is hard coded in rally-ovs. If we want to try
different binding strategy, we need to add a new parameter to the rally
task file.