Richard Low of Acunu presented some performance results demonstrating how marvelous Acunu's distribution of Cassandra is compared to the vanilla version. There's a related blog post with updated range query results here: http://www.acunu.com/2011/03/cassandra-under-heavy-write-load-part-ii/
18. Insert summary
Vanilla Cassandra Acunu Cassandra
Insert time / hrs 26 18
Max 95th
percentile latency / 37 0.3
s
19. If you only care about write
performance, /dev/null is
great (and web-scale)
20. Range queries
• High insert rate no use if reads are slow
• High insert rate can impact range query
performance
• Test: insert 100 million rows, with same
properties as before
• After inserts, immediately perform small
range queries (get_range_slices with
random start key)
21.
22. Summary...
• Strong foundations for Big Data with order of
magnitude performance improvement
• v1: Back Cassandra (+others) and tools without
modification
• Fast, robust, simple platform
• Makes test/dev cycle more effective
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
Old\nOracle, MySQL, etc; on top of legacy FS, volume mgr, RAID, hardware\nNew\nNoSQL (to get horizontal scalability)\nHardware evolving fast - but not evenly\nStuff in the middle hasn’t changed much\n\n\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n
* performance is not just about speed, but also reliability + predictability ... \n* note the specific version tree, showing live / dev branches\n* monitoring goes from the cluster down to iostat\n* ... and includes management: easy cluster deployment\n* talk about performance here\n