亚信科技

The operator introduces AISWare MDB to replace existing OLTP. The thread pool in AISWare MDB occupies less system resource and has more throughput, compared to usage of a single client connected to a single thread. Collaboration model in AISWare MDB is able to be switched by coding at application level without impacting the core, when switching context. When used together with thread pool, issues caused by thread pool like shaking or dead lock can be resolved and it’s able to support millions of connections. Section based index in AISWare MDB uses B-Link-tree algorithm and performs read and write in parallel. It’s not required to lock entire tree or the path from root to specified node when modifying b-tree, which dramatically improves simultaneous performance of access to section index. Meanwhile, the merging and recycling of nodes tries to minimize memory occupation. Hash indexing in AISWare MDB supports dynamic hash. Hash index space is gradually scaled up when using hash index, so that hash algorithm can always be kept linear. Meanwhile, rehash is applied to avoid table locks. MVCC in AISWare MDB isolates read and write transactions, that read transaction doesn’t necessarily request read lock; in the case of massive simultaneous writes, MVCC algorithm prevails than other algorithms.