Red Hat Cluster Suite by scz11423

VIEWS: 352 PAGES: 29

									Red Hat Cluster Suite
HP User Society / DECUS
17. Mai 2006

Joachim Schröder
Red Hat GmbH
Two Key Industry Trends

   Clustering (scale-out) is happening
    ●
        20% of all servers shipped will be clustered by 2006. - Gartner
    ●
        Linux clusters are growing at a cagr of 44% per year. - IDC
    ●
        30%+ of Red Hat inquiries are about implementing clusters.
    ●
        Clustering software is a $500m+ market today.
   Storage capacity and complexity are growing
    ●
        Storage (capacity) will continue to grow at 40-60% annually. -
        Aberdeen.
    ●
        SAN management software use grew by 36% year over year...
        ... driven "primarily by customer need to support larger and more
        complex storage networks” . - IDC
    ●
        Storage management software is a $5B+ market today.
What if you could...
Clustering ­ motivation
   Reduce TCO by:
    ●
        Consolidate services to fewer systems
    ●
        Simplify management tasks
    ●
        Increase service uptime
   Approaches:
    ●
        Scale­up: few very powerful systems running multiple services
    ●
        Scale­out: many small systems collaborate on several services
   Requirements:
    ●
        Data sharing
    ●
        Monitoring infrastructure
    ●
        Service management layer
Clustering – “ buzzwords”
   Application Failover
    ●   Detect service failures and failover to standby systems
   Load Balancing
    ●   Distribute service load over multiple systems
    ●   Common for network based systems
   Active/Passive
    ●   Active node owns service, passive node is idle
    ●   Failover requires “ hand-over” of resources to passive node
    ●   Advantage: generally any application
   Active/Active
    ●   Multiple nodes are owning the service
    ●   Requires typically data sharing
    ●   Application needs active/active-support
Red Hat Clustering Architecture
   Red Hat Cluster Suite provides
                                                                 Cluster File         Cluster Logical
     ●
         Application failover                                      System            Volume Manager*
     ●
         Improves application availability
                                                               Red Hat Global File System (GFS)
     ●
         Included with GFS
     ●
         Core services for enterprise cluster                                              IP Load
                                                                  HA Services
         configurations*                                             (Failover)           Balancing
   Red Hat Global File System (GFS)                                           Core services*:
                                                               DLM – Connection Manager – Service Manager
     ●
         Cluster-wide concurrent read-write file                I/O Fencing – Heartbeats – Management GUI
         system                                                     Red Hat Cluster Suite
     ●
         Improves cluster availability, scalability
         and performance                                       Single node LVM2
     ●
         Includes Cluster Logical Volume Manager                   Red Hat Enterprise Linux
         (CLVM)*


               * These features will be available with launch of Cluster Suite v4 and GFS 
                             6.1 – availability concurrent with RHEL4 Update 1
Red Hat Cluster Suite: Core Cluster
Services
   Core functionality for both Clustering and GFS is delivered in Red Hat Cluster Suite
     ●
         Membership management
     ●
         I/O fencing
     ●
         Lock management
     ●
         Heartbeats
     ●
         Management GUI
   Support for up to 300 nodes
   Two selectable lock management models
     ●
         Client-server with SLM/RLM (single/redundant lock manager)
          ●
              Was in prior version
     ●
         Distributed Lock Manager (DLM)
          ●
              New with Red Hat Cluster Suite v.4
          ●
              Open, stable API – consistent with VMS DLM
Red Hat Cluster Suite Overview
   Provides two major technologies
     ●
         High Availability failover – suitable for unmodified applications
     ●
         IP Load Balancing – enables network server farms to load share IP load
   New with Cluster Suite v4
     ●
         Elimination of requirement for shared storage
          ●
              Significantly reduces the cost of high availability clustering
          ●
              Shared Quorum partition is no longer required
                ●
                  Service state, previously stored in Quorum partition, is now
                  distributed across cluster
     ●
         Online resource management modification
          ●
              Allows services to be updated without shutting down (where possible)
     ●
         Additional fencing agents
Red Hat Cluster Manager Architecture
                      Clients



         Node 0    Network Power Switch
                                          Node 1
          Oracle                            NFS


           bind        Heartbeat




                        SAN
Handling Failures
   Software Failures
     ●
         Monitoring provided by resource agents handles restarting
     ●
         If a resource fails and is correctly restarted, no other action is taken
     ●
         In the event that the resource fails to restart, RHCM will stop and relocate the
         entire service to another node
   Hardware, cluster failures
     ●
         If the cluster infrastructure evicts a node or nodes from the cluster; RHCM
         selects new nodes for the services it was running based on the failover
         domain if one exists
     ●
         If a NIC fails or a cable is pulled (but the node is still a member of the
         cluster), the service will be relocated
   Double Faults – Usually difficult or impossible to choose a universally correct
    course of action when one occurs. Ex: Node with iLO losing all power vs. pulling
    all of its network cables.
After a node failure...
                       Clients



          Node 0    Network Power Switch
                                           Node 1
                                             NFS
                                            Oracle
                        Heartbeat            bind




                          SAN
Data Sharing
   Purpose: parallel access to data resources
    ●
        Files and Directories
    ●
        Needed identical or single data

   Data Sharing technologies
    ●
        SAN – block based sharing
    ●
        NAS – file based sharing
    ●
        Data Replication on cluster nodes
         ●
             File or block based, typically not symmetric
Red Hat Global File System v6.1
   New version for Red Hat Enterprise Linux v.4
     ●
         Uses new common cluster infrastructure in Red Hat Cluster Suite (included)
   Provides two major technologies
     ●
         GFS cluster file system – concurrent file system access for database, web
         serving, NFS file serving, HPC, etc. environments
     ●
         CLVM cluster logical volume manager
   Fully POSIX compliant
   Much faster fsck
     ●
         Ported to GFS 6.0 on RHEL 3 Update 5
   Data and meta-data journaling (per-node journals, clusterwide recovery)
   Maximum filesize & file system size: 16TB with 32-bit systems, 8EB with 64-bit
    systems
   Supports file system expansion
   Supports SAN, iSCSI, GNBD
   Default use of new Distributed Lock Manager (DLM)
     ●
         Client-server locking optional
Distributed Lock Manager
   Red Hat Cluster Suite v.4 includes a Distributed Lock Manager (DLM)
     ●
         Primarily used by Red Hat Global File System
     ●
         Available for general purpose use by any application
     ●
         Closely mirrors the original Digital VMS DLM
   A DLM is a highly functional, distributed (cluster-wide), application
    synchronization subsystem
     ●
         Processes use the DLM to synchronize access to a shared resource (e.g. a
         file, program, or device) by establishing locks on named resources
     ●
         Permits the creation of distributed applications
          ●
              e.g Oracle RAC (which uses a private DLM)
     ●
         Provides a collection of services
          ●
              Multiple lock spaces and concurrency (lock) modes
          ●
              Lock hierarchies/domains (resources & subresources)
          ●
              Range locking
          ●
              Lock conversions & value blocks
          ●
              Blocking notifications
          ●
              Asynchronous completions
Cluster Logical Volume Manager (CLVM)
   CLVM builds upon LVM 2.0 and the kernel device mapper component
    included in 2.5 and 2.6 Linux distributions
   Essentially a cluster-aware version of LVM 2.x
   Commands, features, functions all work just fine in a cluster, any Linux
    server may mount any volume
   Provides
     ●
         Cluster safe volume operations
     ●
         Cluster-wide concatenation and stripping of volumes
     ●
         Dynamic Volume resizing
     ●
         Cluster-wide snapshots (mid '06)
     ●
         Cluster-wide mirroring (mid '06)
     ●
         Other RAID levels (mid '06)
Use Case: High availability with local data
   Red Hat Enterprise Linux (ext3) + Red Hat Cluster Suite
    v4
   Requires NO ADDITIONAL HARDWARE                                    Local storage only
     ●
         No physically shared storage                                  Application failover
          ●
              Prior to RHEL4, physically shared storage was
              required
   Ensures that an application stays running
     ●
         Monitors all cluster nodes
     ●
         Fails-over applications/services from stopped nodes             H/A Cluster
     ●
         Simple management GUI
     ●
         Service definitions are automatically propagated and
         synchronized, cluster-wide
                                        Applications:
                Topology:                                       Data model:
                                        - Read-only data
                - 2 or mode nodes                               - Unshared, static data
                                        - Small web serving
                - NIC card                                      - Replication w/ rsync
                                        - NFS/FTP serving
                - NO shared storage                             - ext3 file system
                                        - Edge of network
Use Case: Shared file datastore, NAS
   Red Hat Enterprise Linux (NFS)           Shared storage – file based (NFS)
   Uses an external NFS file server          Cluster nodes are NFS clients

           Addnl. Cluster features:
           - Increased storage capacity
           - Applications can failover and
           access the same data as before
           - NFS sharing model                                       NetApp, etc.
                                                      HA Cluster
            Applications:
            - Small/medium web serving
            - NFS/FTP serving
            - Read/write data
            - File level access               HPC Cluster

              Data model:
              - Physically shared data
              - File based (NFS)
                                                NetApp, etc.
Use Case: Shared block datastore, SAN
   Red Hat Enterprise Linux (ext3, integrated
                                                  Shared storage – block based (SAN)
    HBA drivers)
                                                 Partitions accessed on a per node basis
   Offers higher, SAN-based, performance
             Addnl. Cluster features:
             - SAN access provides:
             -- improved performance
             -- heterogeneous access
             by other systems

            Applications:
            - Medium/large web serving
            - Medium database                                     SAN
            - Read/write data
            - Block level access

            Data models:
            - Storage Area Network
            - Physically shared storage
            - Block based
            - ext3 file system
Use Case: Concurrent file system access
   Red Hat Enterprise Linux + Red Hat Global
                                                       Shared storage – block based (SAN)
    File System
                                                   Partitions concurrently accessed by all nodes
     ●
         Includes Red Hat Cluster Suite
   Shared file system access
   Scalable performance
   Same hardware as previous configurations
   Common software infrastructure with
    previous configuration

                                                                      SAN


          Applications:
                                 Data models:                     Addnl. Cluster features:
          - Large web serving
                                 - Storage Area Network           - SAN access provides:
          - Large database
                                 - Shared file system access      -- improved performance
          - Read/write data
                                 - Block based                    -- heterogeneous access
          - Block level access
                                 - GFS file system                by other systems
          - Concurrent access
About Piranha & IPVS
   Piranha
     ●
         Front-end to IPVS (or LVS) which enables IPVS director failover
     ●
         Provides a web-based configuration GUI
   IPVS – does the heavy lifting
     ●
         Layer-4 (application) switching/routing
     ●
         Multiple scheduling algorithms: round-robin, weighted round-robin, least-
         connections, weighted-least-connections, and more...
     ●
         Several routing topologies: NAT, Direct Routing, IP Tunneling
          ●
              DR and TUN are tricky to set up, and require configuration on the Real
              Servers
Piranha Architecture
   Appears as one mega-machine to clients: SSI at the IP level
   One or two nodes acting as IPVS directors
     ●
         Stand-alone cluster: Not dependent on CMAN, GULM, etc.
     ●
         IPVS routing table is replicated between the primary and backup directors for
         hot-standby operation
     ●
         No dependencies on shared storage (no shared data to protect)
   Real servers can use GFS, NFS, or manual replication (using rsync, for
    example)for static content
   Provides high availability for applications via redundant Real Servers and
    (optionally) a redundant director
Piranha Architecture

  Director
                          Internet
  Real Server

                           Virtual IP




                Primary    Heartbeat    Backup




                   RS0     RS1          RS2
Piranha – Real Server Failure

  Director
                          Internet
  Real Server

                           Virtual IP




                Primary    Heartbeat    Backup




                   RS0     RS1          RS2
Piranha – Director Failure
                          Internet
  Director


  Real Server

                           Virtual IP




                Primary    Heartbeat    Backup




                   RS0     RS1          RS2
                    New!

    Red Hat and GFS Joint Linux Clustering Solution
   HP Serviceguard for Linux creates a cluster of Linux servers that make 
    application services available despite hardware or software failures or 
    planned downtime.
   Red Hat Global File System (GFS) is a cluster file system that enables 
    multiple servers to simultaneously read and write to a single shared file 
    system on a Storage Area Network.
   Recently tested. Working together, now provide best of breed clustering 
    technologies for Linux. The combination offers greater availability of 
    applications and data using a cluster of low­cost Linux servers.
     ●
         White Paper 
     ●
         Solution Brief 
     ●
         Joint marketing
Questions?
Vielen Dank!



Joachim Schröder, Solution Architect
joachim.schroeder@redhat.com

								
To top