More and more SMEs are transferring their former local workloads to cloud infrastructure. During this process many things change, from network speed and traffic to writing capacity of underlying hardware. Despite the possibilities that cloud computing offers, one has to be aware of the bottlenecks to avoid and overcome them.

During the development of the workflow, various limitations that have never played a role on local computing infrastructure now have to be taken into account. Normally there used to be only a limited amount of parallel processes on each local machine, making things like read/write performance much less important than dealing with 300 parallel Kubernetes jobs trying to write to one attached disk. This limitation can be avoided using other storage classes resulting in higher costs, otherwise computing performance might drop significantly. Another factor is bandwidth that within local networks only seldom limits the transfer rates, while this might be the case transferring results from cloud infrastructures back to local hardware. Transferring a local workflow to a cloud cluster might sometimes feel as going from bottleneck to bottleneck, while most of the bottlenecks only arise because computing power and at the same time possibilities rise into unknown spheres.