April 2016 Summaries

A Gremlin traversal is an abstract description of a legal path through a graph. It involves the creation and branching of traversers, which are pointers that move through the graph according to instructions dictated by the traversal. Traversers can die along the way due to filtering or walking down dead-end subgraphs. The surviving traversers provide answers based on their lineage across the graph. Gremlin OLAP is a distributed computing model for executing Gremlin traversals against any TinkerPop-enabled graph system, using a message passing algorithm. It involves vertex programs that are evaluated at each vertex in parallel and a master traversal that coordinates the computation. Some steps require aggregation at the master traversal, while others involve reducing barriers to yield an aggregate of their reduced associative/commutative form. The TraversalVertexProgram drives the evaluation of distributed traversals, with the complexity of the computation residing in both the vendor's GraphComputer implementation and Apache TinkerPop's Traversal implementation.

This article discusses performance improvements made to Apache Cassandra's COPY FROM command using profiling tools like cProfile and line_profiler. The changes introduced by CASSANDRA-11053 optimized the process, increasing performance from around 35,000 rows per second to as much as 117,000 rows per second. The optimization involved introducing a feeder process for reading data and moving csv decoding to worker processes. Additionally, replacing the queue with a pool of pipes improved communication across processes. Python performance recommendations were also applied, such as using built-in Integers instead of Python types and storing function references in local variables before entering loops. The final performance results varied depending on factors like CPU scheduling and data type complexity.

This article discusses methods for improving performance when importing CSV data into Cassandra via cqlsh COPY FROM. Key suggestions include installing a Python driver with Cython and libev C extensions, setting the CQLSH_NO_BUNDLED environment variable to use an installed driver, compiling the clqsh copy module with Cython for additional performance boosts, adjusting CPU scheduling on Linux systems, and experimenting with six COPY FROM parameters to optimize performance for specific workloads. These methods can result in significant improvements in import speed and efficiency.

Duy Hai Doan, an Apache Cassandra Evangelist at DataStax, discusses the technical implementation of native secondary index in Cassandra and highlights best use-cases and worst anti-patterns. The article explains how secondary index is a distributed index that stores index data alongside original data on the same node to avoid issues with consistency levels and wide partitions. It also delves into the design schema for inverse index, read path optimization, and common anti-patterns to avoid when using native secondary index.