Cassandra Performance
Past, Present & Future
Tom Wilkie, FounderAcunu Ltd
11 Jan : Cassandra 0.7.0
3 Jun: Cassandra 0.8.0
18 Oct: Cassandra 1.0.0
A Busy Year
11 Jan : Cassandra 0.7.0
• Dynamic endpoint snitch
• Row cache improvements
• Secondary indexes
3 Jun: Cassandra 0.8.0
18 Oct: Cassandra 1.0.0
11 Jan : Cassandra 0.7.0
3 Jun: Cassandra 0.8.0
• Counters
• CQL
18 Oct: Cassandra 1.0.0
11 Jan : Cassandra 0.7.0
3 Jun: Cassandra 0.8.0
18 Oct: Cassandra 1.0.0
• Compression
• Off heap caches, arena allocations etc
• LevelDB-style compactions
Benchmarking
http://www.oracle.com/technetwork/database/nosqldb/learnmore/nosql-database-498041.pdf
Read the fine print
24GB of memory ... a 300GB disk
...key size of 13 bytes and data size of 1108 bytes.
... 96 nodes storing 2.1 billion records
~350k inserts / sec
“Read/write performance on a single four-core machine. One million inserts followed by one million updates”
http://www.datastax.com/dev/blog/whats-new-in-cassandra-1-0-performance
Machine1 x Quad core Xeon E562024GB RAM, 8x1TB Disks
1 commit log, 7 in RAID 10
Workload
Keys: 8 byte, Values: 10 bytesRandom key distribution
10 million Inserts,Random Point Gets,
Range Queries
0
10,000
20,000
30,000
40,000
50,000
60,000
70,000
80,000
90,000
Inserts Point Gets Range Queries
0.7.90.8.71.0.2
Cassandra Performance Evolution
Machine1 x Quad core Xeon E562024GB RAM, 8x1TB Disks
1 commit log, 7 in RAID 10
Workload
Keys: 8 byte, Values: 100 bytesRandom key distribution
300 million Inserts,Random Point Gets,
Range Queries
Time / hours
Inse
rts/
s
0
10,000
20,000
30,000
40,000
50,000
60,000
Inserts
1.0.2w/ Compressionw/ Leveled Compaction
Effect of Compression and Leveled Compaction
0
40
80
120
160
200
Point Gets
0
10,000
20,000
30,000
40,000
50,000
60,000
Inserts1.0.2w/ Compressionw/ Leveled CompactionAcunu
Effect of Compression and Leveled Compaction
0
100
200
300
400
500
Point Gets Range Queries
Future
Questions?Tom Wilkie@tom_wilkie
http://bitbucket.org/acunuhttp://github.com/acunu
http://www.acunu.com/downloadhttp://www.acunu.com/insights
Recommended