Citation Context

...may involve a global graph to control conflicts [6, 11], or atomic broadcast to order transactions [27, 37]. Later work considers snapshot isolation as a more efficient alternative to serializability =-=[15, 17, 18, 30, 39, 52]-=-. Walter differs from the above works because they ensure a stronger isolation property—serializability or snapshot isolation—which inherently requires coordination across sites to commit, whereas Wal...

Citation Context

...ies. Kemme et al. [21] have discussed how to implement different isolation levels (including serializability) in replicated databases using group communication primitives. Two systems, Postgres-R(SI) =-=[31]-=- and Ganymed [24], have been built to replicate databases that use snapshot isolation. Fekete et al. [11, 13] have studied the conditions under which transactions executing on a centralized database w...

Citation Context

...apshot isolation) in replicated databases using group communication primitives. In addition, they implement Postgres-R [11], and integrate the replica management with the database concurrency control =-=[24, 14]-=- in Postgres-R(SI). Postgres-R(SI) uses a replicated form of snapshot isolation. In contrast to Postgres-R(SI) [24, 14], the two Tashkent systems have the key feature of uniting ordering and durabilit...

Citation Context

...modifications. This paper is related to all previous work studying transparent performance optimizations in static or dynamic content servers, such as web caching [32, 2], scaling through replication =-=[5, 35, 21, 22, 38, 39, 31]-=-, load balancing [28, 7, 25] or admission control [37, 12]. In contrast, we present our experience with scaling a dynamic content server system, as a whole. We strive to provide insights into what tec...

Citation Context

...stency models instead of 1-copy serializability. A related trend is to support levels of consistency that are generally considered acceptable for transactional applications, such as snapshotisolation =-=[32, 21]-=- and session consistency [13], or a continuum of consistency models with tunable parameters [33, 25]. On the other hand, for some applications, adjusting the application to a weaker consistency model ...

Citation Context

...er of aborts [1, 2, 15, 22, 23, 24, 25, 27]. Some approaches have also considered consistency criteria weaker than serializability. In particular, many recent works have focused on snapshot isolation =-=[5, 6, 18, 26, 29, 33]-=-. Sprint “decouples” replication for availability from replication for performance. Data servers can be replicated for performance only: if a data item is often read and rarely updated, then increasin...

Citation Context

...This is in fact what guarantees that the history restricted to update transactions generated by GMU is 1CS, as demanded by US. However, unlike existing distributed/replicated MVCC schemes [21], [22], =-=[27]-=-, [13], GMU does not order transaction commit events by relying on a centralized, or fully replicated, global clock. Conversely, GMU relies on a novel, highly scalable, fully distributed synchronizati...

Citation Context

... 9]. The traditional correctness criterion for data replication is one-copy-serializability (1CS) [4]. More recently, some protocols have explored extensions of snapshot isolation for replicated data =-=[7, 23, 27, 16, 8]-=-. All the protocols on partial replication implement 1CS and most of them assume that all data accessed by a transaction is stored on the site where the transaction is submitted. One of the earliest p...