Database technology for embedded software systems

商业关系引擎是一个价值数十亿美元的产业，并且证明了开发和维护信息系统的成本降低，这些系统源于SQL语言的使用以及这些引擎的后续部署。所提议的研究计划是基于这样一个假设，即开发和维护嵌入式软件的成本的相应降低也将随着SQL语言的使用而产生，而这反过来又需要许多新的或改进的底层技术的能力。首先，对于涉及概念数据库模式和内部数据库模式的更丰富形式的情况，需要改进查询优化，因为查询优化器生成的查询计划在性能上必须与专业程序员直接编写的低级代码相当。这对查询优化的各个方面的复杂性提出了更高的要求，包括物理数据模型和数据编码、视图集成功能和概念数据建模。其次，嵌入式系统的许多属性会影响数据的索引和编码。例如，在此设置中选择适当索引的决策现在可以由阈值需求驱动，这些阈值需求规定特定查询必须在对数甚至常数时间内运行。其他后果是需要更通用和可靠的数据集成功能，以及在主存储器或闪存中替代本地数据编码。第三，需要一种可替代的事务模型，在这种模型中，数据库服务器可以在客户端与服务器交互的动态方面更加了解客户端。例如，这可以实现无死锁锁定协议的使用以及并发控制与数据恢复的解耦，这是许多嵌入式系统的基本需求，在这些系统中，日志记录和常规中止功能将导致不可接受的运行时开销。

Commercial relational engines are a multi-billion dollar industry, and are testimony to the reductions in the cost of developing and maintaining information systems that derive from the use of the SQL language and the consequent deployment of these engines. The proposed research program is guided by the hypothesis that comparable reductions in the cost of developing and maintaining embedded software would also ensue from the use of the SQL language, which in turn requires a number of new or improved capabilities with the underlying technology.    First, there is a need for improved query optimization for cases involving much richer forms of both conceptual and internal database schema since query plans generated by query optimizers must be comparible in performance to low-level code written directly by expert programmers. This places much greater demands on the sophistication of various facets of query optimization, including physical data models and data encoding, view integration capabilities and conceptual data modeling.    Second, there are a number of properties of embedded systems that affect the indexing and encoding of data. For example, decisions on the selection of appropriate indexing in this setting can now be driven by threshold requirements dictating that a particular query must run in logarithmic or even constant time. Other consequences are the need for more general and reliable data integration capabilities, and on alternative local data encoding in main or flash memory.    And third, there is a need for alternative transaction models in which a database server can be much more client aware in terms of the dynamic aspects of client interactions with the server. This can enable, for example, the use of deadlock free locking protocols and a consequent decoupling of concurrency control from data recovery, an essential requirement for many embedded systems in which logging and general abort capabilities would otherwise lead to unacceptable runtime overhead.