1/k phase stamping for continuous shared data (extended abstract)

Abstract

Interactive distributed applications are a relatively new class of applications that are enabled by sharing continuously evolving data across distributed sites (and users). The characteristics of application data include very fine-grained updates that can atomically access a subset of the shared data, masking of update effects, and irregular locality and contention for access. Existing programming approaches are not appropriate for programming such continuous shared data in a wide-area environment.We define an object-set abstraction, where a set is replicated at sites interested in the objects, and is modified using add, delete and update operations. The key features of this abstraction are issue-time access information for update operations, and the potential for replicating the computation associated with updates. Ordering of operations is an important problem, and we present a fast, scalable ordering algorithm, 1/k phase stamping, that uses distributed tokens that correspond to objects in the set. This algorithm provides substantially better performance than alternative algorithms, is deadlock and abort free, and requires no queuing at the tokens. Therefore, tokens can be located inside a programmable 'active' network. The precise stamps generated with 1/k phase stamping enable a dynamic communication optimization technique, effect and stamp merging, which can reduce communication and allow utilization of best-effort message delivery. These algorithms have been implemented in the RAGA system which can tolerate crash failures.

Unconscious Eventual Consistency with Gossips

“…A new class of so-called interactive distributed applications is emerging: distributed virtual environments, interactively steered scientific applications, collaborative design systems, etc [3]. These applications may need to run in a wide area asynchronous environment with widely distributed users and resources and no central authority.…”

“…For space reasons, the snRequest() function is not described. This function simply requests a sequence number from one member of the coalition pi relies on 3. For space reasons, we do not provide the pseudo-code of this mechanism.…”

Normality versus system mobility

“…H ) be a history of S. Let r(x)v an atomic read in a method of H. r(x)v is legal iff the two following conditions hold: (1) objects in X such that objects( ) T objects( ) 6 = . is order before in the object order, also written !…”

“…rw ), defined as follows: Definition 11 (Logical read) Let S = ( P X) be a shared memory system and let H = (H ! H ) be a history of S. Let , and three methods which interfere 2 One could say that it will be interesting to treat each method as a black box -as the o-set defined in [1] -modeling the objects spanned by it as an aggregate object. In the sequel we show that it is impossible to design a normal system using only local information.…”

“…The work of [1] motivates from the practical point of view the use of multi-objects (refereed as o-sets). They show that the complexity of the interactive distributed applications -applications that are enabled by sharing continuously evolving data across distributed sites and users -imposes the use of complex data structures.…”