COSAR-CQN: Enabling Data Intensive COSAR-CQN: Enabling Data Intensive Applications Applications Jason Yap and ShahramGhandeharizadeh Dept. Of Computer Science, University Of Southern California Dept. Of Computer Science, University Of Southern California Introduction Motivation Introduction “Classic” Web Application Motivation • Why? “Classic” Web Application Architecture Cache-Extended Architecture • Why? Steady-state Throughput 90000 …. Users …. Users 50000 60000 70000 80000 r minute …. HTTP …. HTTP 20000 30000 40000 50000 Requests per Load Balancer Load Balancer 0 10000 COSAR-CQN Classic Web Servers SQL Web Servers SQL TCP/ UDP • Challenge: Keep the cache consistent Web Servers Web Servers … DBMS Server Cache Servers DBMS Server Memcached COSAR-CQN Architecture COSAR-CQN Architecture Conceptual architecture (as viewed by an Application COSAR-CQN Advantages: Conceptual architecture (as viewed by an Application Programmer) compared to the underlying Physical architecture: Advantages: • Developer populates the cache without authoring software for cache consistency. architecture: Conceptual Physical software for cache consistency. • Reduced software development cycle: design, development, testing, maintenance. • Faster feature development at reduced costs. • Comparable performance to developer provided solutions How? How? • SQL • Continuous Query Notification (CQN) • Continuous Query Notification (CQN) Examples: Oracle 11g, Microsoft SQL Server 2005, 2008, 2008 R2 2008 R2 Future Research Experimental Results • Predictive Framework: Registration of queries carries its own overhead, so there are cases Measured Request Rate • COSAR-CQN outperforms carries its own overhead, so there are cases where selective registration and caching may enhance overall performance. • COSAR-CQN outperforms Classic even when starting with an empty cache, with enhance overall performance. • Application of COSAR-CQN to RAYS. • Public domain software release of COSAR-CQN. with an empty cache, with an execution time of 1.3 hours vs. 5.2 hours • Public domain software release of COSAR-CQN. • Scalability with cache replication and partitioning. 1.3 hours vs. 5.2 hours for Classic. partitioning. • Monitoring Tools. • Comparison of COSAR with memcached: 2500000 Cache Size Exponential Cost Model • Cache-enabled Database Management Systems. 1000000 1500000 2000000 red Cost (MB) (Seconds) COSAR memcached 32 1,796,640 (19.1%) 2,138,940 64 1,452,450 (33.7%) 1,941,570 0 500000 1000000 Incurr 64 1,452,450 (33.7%) 1,941,570 128 997,470 (67.9%) 1,674,680 256 582,540 (127.9%) 1,327,490 512 400,323 (111.7%) 847,601 DBMS today Cache-enabled DBMS 32 64 128 256 512 1024 1536 Memory Size (MB) 1024 85,110 (66.1%) 141,399 1536 29,268 (-0.1%) 29,244 Sponsored by