White Papers

10 Dell Storage PS Series Arrays: Scalability and Growth in Virtual Environments | TR1072
5 Group and pool transparency
When it comes to scaling PS arrays, there are two trains of thought: scale up or scale out. The PS Series
architecture uses a scale-out architecture, but there are times when scaling out is not enough.
For example, customer A currently has 3 PS arrays and is purchasing 3 more. Looking at customer A’s
storage group, they have 3 arrays and 50 volumes deployed in a single storage pool. The new arrays arrive
and customer A decides to add the 3 arrays to the same storage pool because they will gain load balancing
benefits. Customer A would gain additional load balancing by doing this, but they are already consuming 400
iSCSI connections in the current pool. Instead of placing the 3 arrays in the same pool, it may make more
sense to deploy a new storage pool, place the arrays in the new pool, and start creating volumes out of the
new pool. This will allow customer A to start at 0 with the connection count in the new pool. Fortunately, for
customer A, they can easily create the new pool and perform an online migration of the 3 new arrays from the
current pool to the new pool without disrupting the other volumes or arrays in the current pool.
Operating systems do not distinguish between pools as long as the volume is given access to the server. The
only difference between using a single pool or many pools is the load balancing or performance tier
characteristics of the volume. A volume cannot load balance across storage pools, so to take advantage of
PS Series multimember load balancing for a specific volume, there must be more than one array in the pool
where the volume resides.
Another option is creating a new PS Series Group. By splitting up arrays into 2 groups, connection availability
doubles, but there are some caveats to this approach in some environments. With a new group, a new
dynamic discovery address has to be added to host iSCSI initiators and restrictions such as SAN copy offload
or online volume migration are not available between groups. These limitations can be mitigated with some
thought and proper design, such as using cluster systems that only connect to a single PS Series Group.