quasardb: right to data
- 1. Right to data March 2016 1
- 2. OUR MISSION: GIVE INSTANT ACCESS TO THE RELEVANT DATA 2
- 4. The “direct to data” quasardb technology 4 Fast and frugal ACID transactions Scale-out / Scale-up Automatic replication Batch processing Streaming Unlimited tagging Multi-platform Spark integration And so much more!
- 6. In real life Even for large entries Even under pressure We chose the impossible 6 Reliable Fast Scalable Multi-processor friendly Multi-nodes friendly No software limit Fail gracefully No “small print”
- 7. Don’t search, tag! 7 • Indexing the whole content is slow • Document oriented approach does not scale • Relations don’t scale Our solution: tagging • Scalable • Inexpensive • Simple • Flexible
- 8. Scalable thanks to the Chord Algorithm 8 100 150 200 12 1 • Consistent hashing distribution • Fully symmetric: no master node • Each node only knows about its neighbors • Lookup: O (log n) complexity Crawl the ring in O (log n) to find SHA-3 (key) successor Data Key SHA-3 21 28 1 12
- 9. What reliable means 9 Persistence Distribution Networking Memory Synchronous Consistent Replication Hot plug’n’play Timeout Retries Safe marshaling Greedy Overcome low-memory conditions
- 10. Symmetric, Asynchronous, Atomic and Lock-Free 10 Unlimited parallel read access Highly optimized memory management Elastic and transparent Continuous self- optimization BA No global lock
- 11. Optimized Memory Usage : Zero Copy 11 AFinal result MarshallingA BufferingA A Unmarshalling A Network copy Network transmissionA A Original A A
- 12. 12 To sum up
- 13. Benefits 13 Easy installation and set-up • Simple user interface • Data agnostic • Hot plug’n’play Scalability • Theoretical infinite scalability thanks to a P2P design • Automatic sharding on membership changes • Schemaless design Protect your investment • Complete your ecosystem but does not replace it • Works on all infrastructure types • Interfaces with all market’s standards
- 14. Benefits cont… 14 Reliability • Masterless design ensures fault tolerance system • No single point of failure • No errors even under heavy loads • Data replication provides high availability Frugality • Run on supercomputer and on limited resources connected objects as well • Benchmarks have shown a 40% reduction of computer resources • Consumes only required computer resources
- 16. Roadmap 16 Quasardb 2.5 – Q2 2016 • Time series • Speed optimizations • More APIs • Excel plugin Quasardb 3.0 – Later in 2016 • Fine grained access control • Secure connection • Server-side processing • Advanced tag queries
- 17. 17 What our customers do with quasardb
- 18. Mail sorting systems 18 A couple of milliseconds to perform hundreds of queries. Speed and accuracy of quasardb delivers.
- 19. Document archival 19 Customer dissatisfied with elastic search. Speed, reliability and complexity issues. With quasardb: no more pains points and double the performance!
- 20. Historical financial data 20 Load the terabytes of data you need in a couple of minutes in a big memory machine. And then… Sub-second computations. This is how you should do Big Data.
- 21. 21 How fast is fast?
- 22. Cisco UCS Benchmark 22 0 10 20 30 40 50 60 0 100000 200000 300000 400000 500000 600000 700000 800000 900000 1 node 2 nodes 4 nodes 7 nodes Gigabit/s Transactionspersecond Bandwidth Tps
Editor's Notes
- #7: Crawl the ring in O (log n) to find SHA-3 (key) successor Consistent hashing distribution Fully symmetric: no master node Each node only knows about its neighbors Lookup: O (log n) complexity
- #9: Crawl the ring in O (log n) to find SHA-3 (key) successor Consistent hashing distribution Fully symmetric: no master node Each node only knows about its neighbors Lookup: O (log n) complexity
- #11: Asynchronous I/O Low-level memory management Lockfree containers Transactional memory Assembly optimizations
- #23: Serveur de simulation analytique : Cisco UCS C24M4 - 2 CPU E5-2697 v3 - 256Go -2 x40 Gb VIC 1385 Commutateur : Cisco Nexus 9372 PQ (2 connexions 40 Gbits vers le serveur de simulation – 8 connexions 10 Gbits vers deux fabrics interconnects ) Fabric interconnect : 2 x Cisco UCS 6248 équipés du logiciel UCS manager 1 cluster de 7 nœuds quasardb : chaque nœud est constitué d’un serveur Cisco UCS C240M4 - 2 CPU E5-2697 v3 - 256Go - 2 x 10Gb VIC - 24 DD 1To