Quasi-consistency and Caching with Broadcast Disks [chapter]

Rashmi Srinivasa, Sang H. Son
2001 Lecture Notes in Computer Science  
The challenges ensuing from the asymmetric communication capabilities of mobile environments have led to an increased interest in broadcast-based data dissemination. Among the concurrency control (CC) techniques for transactional clients in broadcast environments, BCC-TI has been shown to be more efficient than a traditional technique [1] . We propose two ways of improving CC performance in broadcast environments: caching and a weaker consistency criterion. We demonstrate that caching improves
more » ... uery response time in BCC-TI. We propose a new CC technique called Quasi-TI that enforces a correctness criterion called quasi-consistency [2]useful when serializability is too expensive to enforce. We introduce a new caching scheme (PIT) and study its effects on Quasi-TI's performance. Through simulation, we demonstrate the benefits of the proposed techniques. Related Work Broadcast-based data dissemination has been studied extensively over a few years [3, 12, 14] . In a broadcast disk model, the server broadcasts all objects in its database in a broadcast cycle and the cycle is executed repeatedly. Clients can read the values of the objects as they are broadcast. The model can be that of a flat disk or of multiple disks. The multiple disk model has been extended to allow updates at the server [12] . Typical applications in broadcast environments have clients that execute read-only transactions (queries), and a server that executes update transactions. Datacycle [13] supports transactions, guaranteeing serializability. CC techniques that exploit the semantics of read-only transactions are proposed in [1, 15] . In [15] , different queries can observe different orders of update transactions. Quasi-consistency differs from this correctness criterion in that it allows a query to specify the amount and type of imprecision it can tolerate. The BCC-TI [1] scheme guarantees serializability using dynamic adjustment of timestamp intervals [16] , and works as follows. During each broadcast cycle, the server stores information on update transactions that committed in that cycle, building a control information table (CIT). The server broadcasts the CIT in the next cycle. At clients, every query has a timestamp interval that is used to record the temporary serialization order induced during execution. When a query reads an object or a CIT, its timestamp interval is adjusted to reflect dependencies between the query and committed update transactions. If the interval becomes invalid (lower bound ≥ upper bound), a non-serializable execution is detected and the query is aborted. BCC-TI has been compared to optimistic CC with forward validation adapted to broadcast environments, and performance gains have been shown [1]. Broadcast methods for queries in [3, 17] do not consider real-time constraints. Semantic-based consistency criteria that are weaker than serializability are presented in [2] , in the context of caching data at clients in information retrieval systems in order to improve query response time. In order to reduce the overhead of keeping cache copies consistent, applications allow cache copies to diverge from the central server copy in a controlled manner. Quasi-caching is a technique that allows such controlled divergence. Examples of quasi-caching constraint types include: 1. arithmetic: based on difference in values of cache copy and central copy. 2. version: based on number of changes between cached and central copies. 3. delay: based on time by which cache copy lags behind central copy. Quasi-caching would be appropriate for a stock trading application. A user interested in stock prices of chemical companies may be satisfied if she reads prices that are within 5% of the true prices (arithmetic constraint). On an update, the server decides whether to propagate the value, based on the client's tolerable imprecision level. Propagation-overhead is cut and flexibility in scheduling propagation increases.
doi:10.1007/3-540-44498-x_11 fatcat:npjr4nad2bacpi6zl7wk7eyq4i