1. Scalability: Citus is designed to scale horizontally, allowing you to distribute your data across multiple machines for faster query performance. On the other hand, Scylla is a NoSQL database based on Apache Cassandra, which also offers horizontal scalability by allowing you to add more nodes to the cluster.

2. Data Model: Citus is a PostgreSQL extension that allows you to scale out your PostgreSQL database by distributing your data. It retains the relational model and supports SQL queries, making it suitable for applications that require complex querying and transactions. On the other hand, Scylla is a wide-column NoSQL database that uses Cassandra Query Language (CQL) for data access. It is optimized for write-heavy workloads and offers high throughput and low latency for large-scale data.

3. Consistency and Availability: Citus ensures strong consistency by managing distributed transactions across nodes in the cluster. It guarantees that all nodes have the most up-to-date data before committing a transaction. Scylla, on the other hand, prioritizes availability over strong consistency. It uses a decentralized architecture based on the Dynamo-style distributed database, which allows for eventual consistency.

4. Architecture: Citus follows a shared-nothing architecture, where each node in the cluster operates independently and communicates through messaging. It shards data across multiple nodes, providing parallel query execution and fault tolerance. Scylla also follows a similar shared-nothing architecture, where each node is self-sufficient and communicates through gossip protocols. It uses a peer-to-peer distribution model to ensure fault tolerance and high availability.

5. Secondary Indexing: Citus supports secondary indexes on distributed tables, allowing you to efficiently query your data based on non-primary key attributes. It automatically synchronizes secondary indexes across nodes, ensuring consistent query results. Scylla, on the other hand, does not support secondary indexes out-of-the-box. Instead, it relies on alternative approaches like materialized views or denormalization to optimize queries.

6. Data Replication: Citus provides automatic data replication for fault tolerance and high availability. It maintains multiple copies of data across different nodes to ensure data durability. Scylla also supports data replication through a technique called replication factor, which allows you to define the number of copies to be stored for each piece of data. This provides fault tolerance and prevents data loss in case of node failures.

In summary, Citus offers horizontal scalability for PostgreSQL databases with support for complex queries and transactions, while Scylla is a NoSQL database based on Cassandra, optimized for write-heavy workloads with high throughput and low latency. Citus focuses on strong consistency and automatic data replication, while Scylla prioritizes availability and eventual consistency. Both databases have their strengths and are suitable for different use cases.