Presto and Impala are both distributed SQL query engines designed for big data environments, however, they have distinct architectures, performance characteristics, and functionality.

Architecture

Presto is an open-source distributed SQL query engine developed by Facebook for running interactive analytic queries against data sources of all sizes, ranging from gigabytes to petabytes. Presto's architecture is based on a federated query engine design, allowing it to query data from multiple sources like Hadoop, S3, MySQL, and Cassandra without data movement. Presto's architecture is comprised of a coordinator node that manages query parsing and execution planning, and multiple worker nodes that execute tasks. This architecture enables Presto to leverage in-memory processing, which is crucial for achieving low-latency responses on large datasets.

Impala was developed by Cloudera and is an open-source massively parallel processing (MPP) SQL query engine that also works with Hadoop data stored in HDFS, Apache HBase, and Apache Kudu. Unlike Presto, Impala employs a daemon process that runs on each data node of the cluster which allows it to optimize performance overall but can limit its scope for interfacing with varied data backends. Impala's architecture enables high performance but at less flexibility in terms of heterogeneous data source integration.

Performance

Presto is generally favored for its processing speed in scenarios involving complex queries over large datasets. Its in-memory processing can outperform Impala, especially where data isn’t residing in HDFS or where multiple data sources are involved in a single query. Also, Presto's ability to execute low-latency queries makes it particularly well-suited for interactive data exploration.

Impala, on the other hand, was specifically designed to bring traditional data warehouse technology into the Hadoop platform, reinforcing its use for large-scale data processing. Impala’s performance is highly competitive when dealing directly with data stored in HDFS, Kudu, or HBase, due to its optimized use of the Hadoop ecosystem and its MPP-style execution which reduces data shuffling and expedites query execution.

Functionality

SQL Support: Presto supports ANSI SQL, including complex queries, aggregations, joins, and window functions. Presto also offers extensions for approximate queries, JSON, and ARRAY data types, which can be beneficial for specific analytical cases.

Impala supports a variant of SQL, like HiveQL, with extensions that are oriented toward distributed queries. It provides robust support for common analytical functions and operators which makes transitioning from traditional SQL environments smoother.

Extensibility and Integrations: Presto can be extended with custom functions, and it supports a wide variety of plugins which allows it to connect with different data sources and to be integrated into various environments.

Impala integrates deeply with the Apache ecosystem, making it ideal for environments where other Hadoop tools and services are in use. It also supports a good set of APIs for integration into third-party BI tools.

Query Functionality: Presto operates on a pull-based model, parsing and planning queries thoroughly before execution. This allows the efficient query planning in diverse environments.

Impala uses a push-based query execution model, which pushes operations down into the Hadoop cluster, hence optimizing the data processing speed especially well-suited for large volumes of HDFS-resident data.

Summary Table

In conclusion, both Presto and Impala bring valuable capabilities to big data querying and analytics. The choice between them should be guided by specific use cases, data environments, and performance considerations. Presto's versatility and speed make it ideal for environments requiring quick, interactive querying across diverse data sources, whereas Impala's robust performance in Hadoop-centric environments makes it suitable for deep analyses where execution time is less sensitive.