cassandra glossary

SuperColumns also differ from regular Columns in that they do not have an associated timestamp.

Chiton
In ancient Greece, a chiton was a cloth garment, typically sleeveless, worn by both men and women. Chiton is the namesake for the open source project Chiton by Brandon Williams, which is a Python GTK-based browser for Apache Cassandra. It is currently hosted at http://github.com/driftx/chiton.
A related project is Telephus, a low-level client API for Cassandra written in Twisted Python. It is currently hosted at http://github.com/driftx/Telephus.

Stage
Part of Cassandra's Staged Event Driven Architecture (SEDA), a stage wraps a basic unit of work. Stages are an important part of how Cassandra can perform well. A single operation may flow between various stages to complete, rather than getting completed in the same thread that started the work.
A stage consists of an incoming event queue, an event handler, and an associated thread pool. Stages are managed by a controller that determines scheduling and thread allocation; Cassandra implements this kind of concurrency model using the thread pool java.util.concurrent.ExecutorService. To see specifically how this works, check out the org.apache.cassandra.concurrent.StageManager class.
The following operations are represented as stages in Cassandra:
Read
Mutation
Gossip
Response
Anti-Entropy
Load Balancer
Migration
Streaming
There are a few additional operations are implemented as stages too, including working with memtables in the ColumnFamilyStore class, and the Consistency Manager is a stage in the StorageService.
An operation may start with one thread, which then hands off the work to another thread, and may hand it off to other threads. This handing-off is not directly between threads, however; it occurs between stages.
Also see SEDA (Staged Event Driven Architecture) (SEDA (Staged Event Driven Architecture)).

SEDA (Staged Event Driven Architecture)
Cassandra employs a Staged Event Driven Architecture to gain massive throughput under highly concurrent conditions. SEDA attempts to overcome the overhead associated with threads. This overhead is due to scheduling, lock contention, and cache misses. The effect of SEDA is that work is not started and completed by the same thread, which can make a more complex code base, but also yield better performance. Therefore, much of the key work in Cassandra, such as Reading, Mutation, Gossiping, memtable flushing, and Compaction, are performed as stages (the "S" in SEDA). A stage is essentially a separated event queue.
As events enter the incoming queue, the event handler supplied by the application is invoked. The controller is capable of dynamically tuning the number of threads allocated to each stage as demand dictates.
The advantages of SEDA are higher concurrency and better management of CPU, disk, and network resources.
You can read more about SEDA as it was originally proposed by Matt Welsh, David Culler, and Eric Brewer at http://www.eecs.harvard.edu/~mdw/proj/seda.
Also see Stage (Stage).

[@more@]

来自 “ ITPUB博客 ” ，链接：http://blog.itpub.net/23937368/viewspace-1051126/，如需转载，请注明出处，否则将追究法律责任。

转载于:http://blog.itpub.net/23937368/viewspace-1051126/