Concurrency Control

Multi-version Concurrency Control (MVCC)

In PieCloudDB, the concurrency control scheme used by each coordinator instance is Multi-Version Concurrency Control (MVCC). This approach ensures that each transaction reads from a consistent snapshot of the database that remains unaltered by concurrent transactions. MVCC offers several advantages over lock-based concurrency control in PieCloudDB, particularly in terms of performance.

Data Model

Relational

PieCloudDB is a rational database, compatible with the complete SQL standard and ACID transactions.

Indexes

Not Supported

PieCloudDB does not support normal index because dense index is extremely expensive for OLAP scenario. However, PieCloudDB uses various data skipping technology to accelerate data access.

Isolation Levels

Read Committed Serializable Snapshot Isolation Repeatable Read

PieCloudDB support two isolation levels: Read Committed and Repeatable Read.

Joins

Nested Loop Join Hash Join Sort-Merge Join Semi Join

PieCloudDB support multiple joins, including Hash Join, Merge Join and Nested Loop Join. And PieCloudDB also support two logical joins: Semi Join and Anti Join.

Query Compilation

JIT Compilation

PieCloudDB is JIT compatible

Query Execution

Tuple-at-a-Time Model

Query Interface

SQL

PieCloudDB is highly compatible with the SQL:2016 standard, fully supporting the SQL:1992 standard, most of the SQL:1999 standard, and some features of the SQL:2003 standard (with a focus on OLAP capabilities). It also supports advanced expressions such as window functions.

Storage Architecture

Disk-oriented

PieCloudDB supports various distributed file system protocol for data storage, including object storage, HDFS and posix. The worker node also uses its local disk as data cache.

Storage Format

Custom

PieCloudDB build its own hybrid column/row (PAX) storage format called "JAMN".

Storage Model

Hybrid

PieCloudDB's horizontal data block into compressed, immutable files, along with the inclusion of footers containing column offsets, provides an efficient and optimized storage system. It improves data compression, accelerates query processing, and simplifies data management within the database.

Stored Procedures

Supported

PieCloudDB support standard store procedure

System Architecture

Shared-Disk

In PieCloudDB, query execution is performed within elastic clusters of virtual machines called virtual warehouses. When executing queries, virtual warehouses utilize related file system protocol to read and write data from distribute file system. The Cloud Service layer in PieCloudDB encompasses a collection of services responsible for managing computation clusters, queries, transactions, and metadata such as database catalogs and access control information.

Views

Virtual Views Materialized Views

PieCloudDB support both materialized views and virtual views.