Tip:
Highlight text to annotate it
X
Guaranteed quality of service isn't simply a feature. It requires an architecture designed
for it from the start.
The fourth requirement for guaranteed quality of service is balanced load distribution.
Let me show you why. In a traditional controller-based architecture, you have a set of disk shelves.
It's very easy for the IO load across those shelves to become unbalanced. Some disk shelves
are lightly loaded. Others have a moderate amount of load. Others are very heavily loaded.
It’s up to the administrator to manually place data to try to keep the IO load balanced
across the system.
When that doesn't occur, you have a problem where additional IO load maxes out one of
the shelves and performance slows to a trickle, even though additional capacity is available
in the system.
Let's look at a SolidFire system here. We got three storage nodes. In the SolidFire
architecture, the data as well as the IO load is distributed evenly across every drive in
the cluster. As you add more nodes with more drives, the capacity and performance is redistributed
evenly across them, resulting in a balanced load distribution. As additional load comes
into the system, that IO load spreads evenly across every drive in the system, resulting
in a continually balanced load distribution and massive throughput as the system scales.
Only with a balanced load distribution can you eliminate hot spots and guarantee quality
of service.