AI-Powered Edge-Cloud Continuum for In-Flight Entertainment and Connectivity Figure 5. Overall CANARIA Cloud IFEC storage system architecture. (a) Cloud IFEC encoding. (b) Cloud IFEC decoding. encoding and decoding. As shown in Figure 5(a), content items are split into smaller segments and encoded into coded segments, which are called symbols. These symbols are then distributed to multiple seats, and each seat receives a different symbol. In order to retrieve a content item, multiple seats will have to be contacted, as indicated in Figure 5(b). Coding ensures that a configurable fraction of the contacted seats can fail, but the content item can still be recovered. The number of symbols per chunk and the number of encoded symbols determines the fault tolerance ofthe content item. The employed coding determines those numbers and can be defined for each content item and each content 22 system individually if the underlying coding scheme offers enough flexibility. Classic schemes for such RAID systems are Reed-Solomon (RS) codes because of their performance guarantees, in the context ofthe cloud IFEC system, rateless codes such as random linear network codes (RLNC) are attractive because oftheir flexibility. The redundancy level of a content item can be chosen depending on its popularity. This is where the storage service has a natural interface to the recommendation system described in the section " AI-Based Data Curation for Passengers. " The recommendation system can provide popularity estimates that can be used to determine the storage location and redundancy level of those items. IEEE A&E SYSTEMS MAGAZINE DECEMBER 2023