This graph compares a serial computer (typical workstation) with a cluster of workstations (similar type). The logarithmic QUIPS is plotted on the y-scale. The cluster has high startup overhead and the overhead increases with the number of the nodes in the cluster. The above graph shows us the scalability of the cluster. It shows that in case an application stresses small subtasks and rapid control changes (i.e it falls in the left hand side of the serial curve) to fit in the cache, it is better to use a single computer.