Avi Kivity of Scylladb deep dives into the internals of Scylladb and what makes it a high performant version of Cassandra, a distributed key-value datastore. Host Nishant Suneja spoke with Avi about how Scylla’s shared nothing architecture helps it in “scaling up” on every single node in the cluster. The discussion further goes into Scylla’s user space IO task scheduler and how does Scylla use the Seastar framework to achieve asynchronous, shared-nothing inter-core communication for IO routing. The discussion also touches upon Scylla’s decision to ditch linux kernel page cache in favor of user-space row-based cache. The show winds up with discussion about Scylla’s cluster level operations and its upcoming feature set in both the community and enterprise editions.
Show Notes
Related Links
- SycallaDB’s shared nothing architecture
- Seastar’s user-space networking stack
- Caching in ScyllaDB
- SE-Radio show on Cassandra
- Scylladb performance benchmarking
- Garbage collection and tail latency
- Java garbage collection
- Seastar framework
- Mirage microkernel
- I/O scheduling in Scylladb
- Guest twitter: https://twitter.com/AviKivity
- Guest email: [email protected]
