A Survey of Cluster Technologies

1. A Survey of Cluster Technologies Ken Moreau Solutions Architect, HP

2. But first, a word from our sponsor… • What this talk will cover • What this talk won’t cover • Political considerations

3. Topics • Single/Multi System View • Shared Root • System Management • Cluster File Systems • Network and Cluster File Systems • Distributed Lock Manager • Configurations • Interconnect • Cluster Alias • Quorum • Application Support • Special coding for clusters? • Failover scripting • Resilience • Data Replication • Disaster Tolerance

4. Accounts Accounts Accounts FC Switch SAN Multi System View Clusters • Systems are relatively independent entities • Disks are physically cabled to multiple systems, but are available to only one system at a time • There is no simultaneous data access from multiple systems: • Provides application failover capability only • The systems look and act differently • The systems are managed independently • This is active-standby Server A Server B

5. Accounts Server C Server D Accounts HR A-M HR N-Z Multi System View Clusters (continued) • Systems are relatively independent entities • Disks are physically cabled to multiple systems, but are available to only one system at a time • There is no simultaneous data access from multiple systems • You can run different applications, or different instances of the same application: federated databases • Provides application failovercapability only • This is active-active FC Switch SAN Server A Server B Inventory Accounts

6. With multi system image clustering Each system disk, whether in a cluster or not, needs to be managed individually Linear increase in management costs and work as environment scales out With single virtual view (SVV) Improved ease of use for single cluster management Synchronization of any group of systems: standalone or clustered Variable levels of synchronization Single Virtual View Two volumes to manage One virtual volume to manage

7. FC Switch SAN HR Payroll HR Server B Server B HR HR Accounts Accounts Accounts Accounts Payroll Single System View Clusters • Systems cooperate very closely • Disks are physically cabled to all systems, and are available to all systems all the time • Simultaneous data access is easy • Provides both application failover and simultaneous execution: shared databases • The systems look and act the same • The cluster is managed as a single entity • Active-active by default Server A Server B

8. Multi or Single System View (1 of 2)

9. Multi or Single System View (2 of 2)

10. Network File Systems I/O Cluster Interconnect Server A Server B Client systems talk toServer systems • Requires 3 I/Os for each disk access • Examples include MSCP, NFS and NTFS FC SwitchSAN Server C Server D Controllers A B C D Isolated Storage

11. Direct Access I/O Cluster Interconnect Direct Access I/O means all nodes in the cluster can talk directly to all disks in the cluster • Provides full transparency and cache coherency • Eliminates 2/3’s of the I/Os in each access to a disk • Only tokens, cache and locks go on the interconnect Server A Server B FC SwitchSAN Server C Server D Controllers A B C D Shared Storage

12. Cluster File Systems (1 of 2)

13. Cluster File Systems (2 of 2)

14. Oracle Support for Cluster File Systems

15. Naming Veritas packages (1) – PA-RISC only (2) PA-RISC and Itanium (3) – 9i RAC and 10g support

16. Extending the HP Serviceguard PortfolioIntegrated bundles with VERITAS software 5.x HP Serviceguard + SGeRAC and VERITAS Storage Foundation™ for Oracle RAC CFS + CVM + DB Accelerator + Mgmt Utilities 7 Available on HP-UX 11i v2 for HP Integrity & 9000 Servers Delivered and supported by HP HP Serviceguard and VERITAS Storage Foundation™ Cluster File System for Oracle (CVM + CFS + DB Accelerator + Mgmt Utilities) HP Serviceguard and VERITAS Storage Foundation™ Cluster File System (CFS + CVM + Mgmt Utilities) 5 6 HP Serviceguard and VERITAS Storage Foundation™ for Oracle Environments (VxVM + FS + DB Accelerator + Mgmt Utilities) HP Serviceguard and VERITAS Storage Foundation™(VxVM, FS, Mgmt Utilities) Premium (+FS, QoS, Fast Mirror ReSync) 1 2 3 4 Oracle Premium(FS, QoS, Fast Mirror ReSync) Standard (VxFS and VxVM) Oracle Standard (ODM, Rollback)

17. MySQL Cluster Architecture • MySQL Servers are the front ends which accept requests and query/update the DB Nodes • DB Nodes (aka “storage nodes”) hold individual copies of the database in memory, respond to the requests from the Servers, and replicate among each other • MGM Node(s) handle system configuration and startup

18. Server A Server B FC Switch SAN A 2-node cluster without a quorum disk • All disks mounted cluster wide • Required quorum = (expected_votes + 2) / 2 = (2+2)/2 = 2 • Actual quorum = (actual_votes + 2) / 2 = (2+2)/2 = 2

19. Server A Server B FC Switch SAN A 2-node cluster without a quorum disk • Servers A and B each try to form a cluster • Actual quorum = (actual_votes + 2) / 2 = (1+2)/2 = 1 • Less than required quorum, so no cluster is formed • What would happen if this scheme wasn’t in place?

20. Server A Server B FC Switch SAN A 2-node cluster with a quorum disk • All disks mounted cluster wide • Required quorum = (expected_votes + 2)/2 = (3+2)/2 = 2 • Actual quorum = (actual_votes + 2)/2 = (3+2)/2 = 2

21. Server A Server B FC Switch SAN A 2-node cluster with a quorum disk • Server A forms a cluster • Actual quorum = (actual_votes + 2)/2 = (2 + 2)/2 = 2 • Server B does not form a cluster • Actual quorum = (actual_votes + 2)/2 = (1 + 2)/2 = 1

22. Cluster Configurations (1 of 2)

23. Cluster Configurations (2 of 2)

24. Break

25. Application Support (1 of 2)

26. Application Support (2 of 2)

27. Resilience (1 of 2)

28. Resilience (2 of 2)

29. Availability Spectrum

30. Boston Nashua Server A Server B Server C Server D Application Servers Application Servers Remote Mirroring DWDM Switches Switches Nashua DuplicateStorage System Nashua ProductionStorage System Boston ProductionStorage System Boston DuplicateStorage System Multi-System View Physical Replication - active/active - data sink - active/passive

31. Network Boston Nashua Server A Server B Server C Server D Application Servers Application Servers VolumeShadowing DWDM Switches Switches ProductionStorage System ProductionStorage System Single-System View Physical Replication

32. Transaction Replication Replicate the transactions themselves • HP NonStop/SQL, HP Reliable Transaction Router, IBM MQSeries and Oracle DataGuard replicate the transactions and stream them to the multiple systems to be executed • Oracle DataGuard “mines” the re-do logs to recreate the transactions to be executed remotely (SQL apply) • The remote database(s) can be used because the operations are coming in as normal transactions, so the replicated databases are not necessarily identical to the original Replay the logs of the effects of the transactions • HP NonStop Kernel AutoSync monitors specified files for changes, and then does whole file replication • HP NonStop Kernel Remote Datacenter Facility (RDF) applies the transaction logs to the remote database(s), at the table or file level • Oracle DataGuard applies the re-do logs directly to the remote databases, at the database level (re-do apply) • The remote database(s) are inaccessible because the replication software requires exclusive access, which forces the databases to stay identical

33. 4 3-6 3-6 3 1 2 2 5 Witness Principal Mirror Application SQL Server SQL Server Log Data Log Data SQL Server 2005 Database Mirroring • Database Failover • Very fast failover,less than 3 seconds • Automatic or manual • Automatic, transparentclient re-direction • Database snapshots • Works with standardserver and storage

34. Disaster Tolerance (1 of 3)

35. Disaster Tolerance (2 of 3)

36. Disaster Tolerance (3 of 3)

37. Summary • Every system offers a high availability option • But the recovery times vary from many minutes to transparent • Every system can scale outside the box • But the sizes vary from 2-node to 255-node clusters • Every system has the option of disaster tolerance • But the technologies vary from one-way data replication between separate clusters, to full active/active cooperation of a single cluster spread over several geographically dispersed datacenters • Understand the options and choose the right technologies • Understand what you get and don’t get with each technology

38. Resources • HACMP • http://www.ibm.com/servers/aix/products/ibmsw/high_avail_network/hacmp.html • Linux LifeKeeper • http://h18000.www1.hp.com/solutions/enterprise/highavailability/linux/index.html • MySQL • Marathon everRun • http://www.marathontechnologies.com/products.html • NSK • http://h71033.www7.hp.com/object/NTLLink.html • OpenVMS Cluster Software • http://h71000.www7.hp.com/openvms/products/clusters/index.html • Oracle 9i RAC, 10g • http://otn.oracle.com/products/database/clustering/index.html

39. Resources • PolyServe Matrix Server • http://www.polyserve.com/products.html • Serviceguard • http://docs.hp.com/hpux/ha/index.html • SQL Server 2000/2005 • http://www.microsoft.com/sql/techinfo/default.asp • SunCluster • http://wwws.sun.com/software/cluster/index.html • TruCluster • http://h30097.www3.hp.com/docs/pub_page/cluster_list.html • Veritas • http://www.veritas.com/van/articles/3245.html • Windows 2000/2003 • http://www.microsoft.com/windows2000/en/datacenter/help

40. Books • “Clusters for High Availability”, Peter Weygant, ISBN 0-13-089355-2 • “In Search of Clusters”, Gregory F. Pfister, ISBN 0-13-899709-8