Download as PDF, PPTX
Uploaded byAntonio Severien
Relational Cloud
The document discusses the 'Relational Cloud,' a database-as-a-service (DBaaS) that addresses challenges such as efficient multi-tenancy, elastic scalability, and database privacy. It outlines technical aspects including resource estimation, graph-based partitioning, and encryption strategies for secure SQL queries. The authors present experimental results that demonstrate the practical viability of DBaaS in cloud environments.
More Related Content
Relational Cloud
- 1. Rela%onal Cloud A Database-‐as-‐a-‐Service for the Cloud Paper by Carlo Curino et al. @mit.edu Presenta%on by Antonio Severien [email protected]
- 2. Overview Ø Rela%onal Databases Ø Database-‐as-‐a-‐Service (DBaaS) Ø Problems AEacked Ø Efficient Mul%-‐tenancy Ø Elas%c Scalability Ø Database Privacy Ø Rela%onal Cloud Ø Experiments Ø Conclusion 2
- 3. Rela%onal Cloud Ø Rela%onal Databases Ø Database-‐as-‐a-‐Service (DBaaS) Ø Problems AEacked Ø Efficient Mul%-‐tenancy Ø Elas%c Scalability Ø Database Privacy Ø Rela%onal Cloud Ø Experiments Ø Conclusion 3
- 4. Rela%onal Databases Ø 1970 by Edgar Codd, IBM research San Jose Ø Tables Ø Rows à Tuples Ø Columns à AEributes Ø Rela%onal Database Management Systems (RDBMS) 4
- 5. Rela%onal Cloud Ø Rela%onal Databases Ø Database-‐as-‐a-‐Service (DBaaS) Ø Problems AEacked Ø Efficient Mul%-‐tenancy Ø Elas%c Scalability Ø Database Privacy Ø Rela%onal Cloud Ø Experiments Ø Conclusion 5
- 6. Database-‐as-‐a-‐Service (DBaaS) Ø Cloud Ø Reduce management, opera%onal and energy costs Ø Elas%city and scale economy amazon RDS Ø Pay-‐per-‐use 6
- 7. Rela%onal Cloud Ø Rela%onal Databases Ø Database-‐as-‐a-‐Service (DBaaS) Ø Problems AEacked Ø Efficient Mul%-‐tenancy Ø Elas%c Scalability Ø Database Privacy Ø Rela%onal Cloud Ø Experiments Ø Conclusion 7
- 8. Problems AEacked Efficient Mul%-‐tenancy Elas%c Scalability Privacy 8
- 9.
- 10. Efficient Mul%-‐tenancy Ø Reduce costs Ø Efficient usage of resources Ø Maximize hardware u%liza%on Ø Single database server on each machine Ø Maintain applica%on query performance 10
- 11. Efficient Mul%-‐tenancy Ø Reduce costs Ø Efficient usage of resources Ø Maximize hardware u%liza%on Ø Single database server on each machine? Ø Maintain applica%on query performance Virtual Machine 11
- 12. Efficient Mul%-‐tenancy Ø Problems Ø Monitoring resource requirements for workloads Ø Predic%ng the load generated Ø Assigning workloads to physical machines Ø Migra%ng workloads between nodes Ø Live migra*on 12
- 13. Efficient Mul%-‐tenancy Ø Kairos (Monitoring and consolida%on engine) Ø Resource Monitor Disk ac%vity and RAM requirements Ø Combined Load Predictor CPU, RAM, Disk model that predicts the combined resource requirements Ø Consolida%on Engine Non-‐linear op%miza%on techniques to… … minimize the number of machines needed … balance load between back-‐end machines 13
- 14.
- 15. Elas%c Scalability Ø Workload exceeds single machine capacity Ø Scale a single database to mul%ple nodes Ø Scale-‐out by query processing par%%oning Ø Granular placement and load balance on backend 15
- 16. Elas%c Scalability Ø Strategy well suited for OLTP and Web workloads… but can extend to OLAP Ø Minimize cross-‐node distributed transac%ons Ø Workload-‐aware par**oner Ø Par%%on data to minimize mul%-‐node transac%ons Ø Front-‐end analyses execu%on traces represented as a graph 16
- 17. Graph Par%%oning we=2 id name age salary id name age salary we=1 we=10 id name age salary we : weight of edge 17
- 18. Graph Par%%oning we=2 id name age salary id name age salary we=1 we=10 id name age salary we : weight of edge 18
- 19. Graph Par%%oning id name age salary id name age salary id name age salary 19
- 20. Privacy 20
- 21. Privacy Ø Adjustable security Ø Onion ring encryp%on design 2 onion layer and 1 homomorphic encryp%on of integer Ø SQL query on encrypted data Ø Security level dynamically adap%ve Converge to an overall security level 21
- 22. Onion Layers of Encryp%on 6. RND: no func%onality 5. RND: no func%onality HOM: addi%on int value 4. DET: equality selec%on 3. OPE: inequality select, min, max, sort, group-‐by 2. DET: equality join or 1. OPE: inequality join String search Value Value string value Strong RND = Randomized Encryp%on (no opera%ons allowed) DET = Determinis%c Encryp%on OPE = Order-‐preserving Encryp%on HOM = Homomorphic Encryp%on (opera%ons over encrypted data) Weak 22
- 23. Rela%onal Cloud Ø Rela%onal Databases Ø Database-‐as-‐a-‐Service (DBaaS) Ø Problems AEacked Ø Efficient Mul%-‐tenancy Ø Elas%c Scalability Ø Database Privacy Ø Rela%onal Cloud Ø Experiments Ø Conclusion 23
- 24. Rela%onal Cloud Architecture Client Nodes Users User Applica%on JDBC-‐client (CryptoDB enabled) Trusted Pla,orm (Private/Secured) Privacy-‐preserving Privacy-‐preserving Untrusted Pla,orm (Public) Queries Results Admin Nodes Frontend Nodes Router Distributed Transac%onal Coordina%on Par%%oning Engine Placement and Migra%on Engine Backend Nodes Backend Nodes Database CryptoDB CryptoDB Par**ons load stats Encryp%on Engine Encryp%on Engine Placement 24
- 25. Rela%onal Cloud Ø Rela%onal Databases Ø Database-‐as-‐a-‐Service (DBaaS) Ø Problems AEacked Ø Efficient Mul%-‐tenancy Ø Elas%c Scalability Ø Database Privacy Ø Rela%onal Cloud Ø Experiments Ø Conclusion 25
- 26. Experiments 26
- 27. Experiments Bad results? Tradeoff for be=er privacy 27
- 28. Experiments Scaling TPC-‐C 28
- 29. Rela%onal Cloud Ø Rela%onal Databases Ø Database-‐as-‐a-‐Service (DBaaS) Ø Problems AEacked Ø Efficient Mul%-‐tenancy Ø Elas%c Scalability Ø Database Privacy Ø Rela%onal Cloud Ø Experiments Ø Conclusion 29
- 30. Conclusion Ø Presented Rela%onal Cloud Ø Efficient Mul%-‐tenancy Ø Novel resource es%ma%on Ø Non-‐linear op%miza%on-‐based consolida%on technique Ø Scalability Ø Graph-‐based par%%oning Ø Privacy Ø Adjustable privacy Ø SQL queries on encrypted data Ø DBaaS is a viable cloud service 30
- 32. References Ø "Rela%onal Cloud: a Database Service for the cloud" Carlo Curino, Evan Jones, Raluca Popa, Nirmesh Malviya, Eugene Wu, Sam Madden, Har Balakrishnan, Nickolai Zeldovich Ø hEp://rela%onalcloud.com 32
- 33. Privacy CryptoDB Example DET-‐encrypted Return to JDBC client decrypted cyphertext RND cyphertexts SELECT i_price, ... FROM item WHERE i_id = N JDBC client decrypts DET level 4 33
Editor's Notes
- #5 Talk about the importance of relational databases and their legacy
- #7 Talk about the market and the viability of relational databases as a service in the cloud
- #11 Make this slide better
- #12 Make this slide better
- #16 Challenge: workload exceeds capacity of single machine
- #17 - THE WAY TO SCALE THE WORKLOADS is to MINIMIZE # of MULTI-NODE TRANSACTIONS… why? OVERHEAD ON HOLDING LOCKS on the BACKEND
- #20 Detail how the provacy works and follow to exemplify on the next sliideKnow well homomrphismUses symetric encryption
- #25 Comparison between consolidated DBs in one machine versus DBs on Virtual Machines.Explained the difference between UNIFORM and SKEWED: uniform load and skewed (50% of the requests goes to one of the 20 DBs)Consolidated 20 databases to one physical machine
- #27 Explain what is TPC-C (benchmarks for databases…. Etc)