Storage-Enabled Adaptive Multi-Path Scheduling Method for Data Transfers in Task Offloading

The delay-sensitive and resource-limited natures of multi-access edge computing (MEC) applications make it necessary to offload local tasks to nearby MEC servers. Large amounts of offloading data are transferred among MEC servers within the stringent deadline constraints. The offloading system hence needs to deliver the data over end-to-end (E2E) connections as soon as they arrive. However, traffic fluctuations make the offloading system difficult to provision the required bandwidth over the E2E connections and guarantee the deadline. In this paper, we present a storage-enabled adaptive multi-path scheduling method (SAM) for offloading transfers across the optical metro network. Instead of delivering the whole data across a single path, SAM can adaptively split the data into small segments and route them through multiple paths. This not only improves the throughput, but also enables large time windows to schedule segments on their own paths within the given deadline. With the large time window, SAM can leverage MEC storage to temporarily store the segment at the intermediate servers when provisioning the entire E2E path fails. Studies show that compared with the state-of-the-art scheduling methods, SAM can accommodate more transfer requests, shorten delivery time and accelerate offloading transfers efficiently. SAM is beneficial to task offloading.