diff --git a/_papers/thesis-report/additional/extended-abstract.typ b/_papers/thesis-report/additional/extended-abstract.typ index e26d023..150da49 100644 --- a/_papers/thesis-report/additional/extended-abstract.typ +++ b/_papers/thesis-report/additional/extended-abstract.typ @@ -119,18 +119,18 @@ The collection of all Solid pods can be interpreted as one big permissioned dece A typical distributed database will both replicate and shard its data @bib1:distributed-database-fundamentals. Sharding data means that the collection of all data is divided into smaller shards, and each machine manages one or more shards. Sharding allows us to scale our data horizontally. -Replication, on the other hand, makes sure that by replicating each shard on multiple machines, the system is partition tolerant~@bib1:cap. +Each shard is then replicated on multiple machines, making the system partition tolerant~@bib1:cap. Different approaches exist to configure the shards and replications. -Often times, each shard will have one leader replications, and the others are followers. -Reads than happen to both leader and followers, while writes only happen to the leader. +Often times, each shard will have one leader replication, and the other replications are followers. +Reads then happen to both leader and followers, while writes only happen to the leader. The leader is responsible for synchronizing all changes to the followers. Such a configuration chooses reads to have eventual consistency~@bib1:distributed-database-fundamentals~@bib1:base, positioning itself on the CAP scale~@bib1:cap @bib1:continous-cap by choosing Availability and Partition Tolerance. The Solid specification builds on top of HTTP and therefore, links can break~@bib1:links-can-break. -This essentially means that there is no partition tolerance, when a pod is disconnected, the data on that pod becomes unavailable. +This essentially means that there is no partition tolerance. When a pod is disconnected, the data on that pod becomes unavailable. Solid thus only has sharding and no replication from a theoretical perspective. -This is an interesting design choice because that means that Solid is jet to position itself on the CAP scale. +This is an interesting design choice because it means that the Solid specification is yet to position itself on the CAP scale. == Concise Bounded Description In this work, we will try to store RDF resources, defined as the CBD (Concise Bounded Description)~@bib1:concise-bounded-description of a Named Node.