There are a number of emerging hardware and software technologies that impact the usage and designs of highly available systems. OpenSAF-based service availability middleware is very complementary to virtualization, multi-core and many other advanced hardware and software technologies. This session will explain the broader technology landscape and will educate on how high availability (HA) fits in.
2. Agenda
Different HA approaches
Application level HA
Virtual Machine Level HA
Virtualization HA Architecture
Integration of OpenSAF in to Virtualization HA
Summary and Recommendations
3. Application HA – Using OpenSAF
OpenSAF Active State Replication
Cluster - App Standby
-SAF Ckpt
Infrastructure - App
Active
Standby
OpenSAF Node
OpenSAF Node local infrastructure
local infrastructure
OS OS
4. Virtualization based HA
Physical Node Active Physical Node Standby
Hypervisor Lock-step Hypervisor
Virtual Virtual
Machine Data- sync Machine
Cluster Mgr Cluster Mgr
5. Virtualization HA architecture
The Active Virtual machine is replicated to the Standby
Virtual machine (Live Migration, Non-live Migration)
Deterministic Replay can be achieved by recording non-
deterministic inputs and then injecting them at the same
execution point during the replay (Lock-step).
If Lock-step interval increases Hypervisor slows down the
Active Virtual machine.
6. OpenSAF in Virtualization HA - 1
Physical Node Active Physical Node Standby
Hypervisor Lock-step Hypervisor
Virtual Virtual
Machine Data- sync Machine
Role Role
PLMC change PLMC change
PLM->CLM->AMF PLM->CLM->AMF
7. OpenSAF in Virtualization HA - 1
The Hypervisors are configured as PLM Execution
environments. Hypervisors integrates with AMF for
determining the Active/Standby
Hypervisor indicates the fault to PLMC or PLMC checks
the health of Hypervisor.
In Fault situation, PLMC sends the information to PLMS,
CLM receives this information from PLMS and node is
declared as out of the cluster. AMF subscribes to CLM
notifications knows this node left the cluster, does the
Active Role assignment to the currently Standby
Hypervisor.
8. OpenSAF in Virtualization HA - 2
Physical Node Active Physical Node Standby
Hypervisor Hypervisor
VM App SAF-CKPT VM App
OpenSAF PLMC OpenSAF PLMC
Payload Payload
PLMC PLMC
Role Role
change change
OpenSAF Controller (PLM->CLM OpenSAF Controller (PLM->CLM
->AMF) ->AMF)
9. OpenSAF in Virtualization HA - 2
In this approach Virtual machines run OpenSAF.
Applications has access to OpenSAF services.
Virtual machines acts as OpenSAF payload nodes.
If the Virtual machine fails the problem gets reported to the
OpenSAF controller and it takes appropriate recovery
actions
If the Hypervisor indicates the fault to PLMC or PLMC
checks the health of Hypervisor.
Based on the Hypervisor capabilities and based on PLM
configuration in imm.xml either Hypervisor and all the VMs
controlled by the Hypervisor are taken out of service or
only Hypervisor is recovered.
11. OpenSAF in Virtualization HA - 3
In this Modeling approach OpenSAF AMF plays significant
role.
The virtual machine and Applications running within the
Virtual machines are monitored through OpenSAF Proxy –
Proxied approach.
The monitoring specifics for the applications and VM are
hidden under Proxy framework.
OpenSAF capabilities are not available for applications
running within VMs.
12. Summary and Recommendations
Virtualization based HA
Live mirroring of VM data is expensive.
It doesn’t work on SMP and multicore environments well.
But this HA environment is non intrusive.
Generally Deployed in 2N redundancy model.
Very Suitable for Legacy application HA.
OpenSAF based HA
Gives lot of utilities and deployment flexibility with redundancy
capabilities (2N, N+M, N-Way, N-Way active, etc).
Transparent to SMP and Uniprocessor environments.
It is intrusive HA and more suitable for new Applications HA.
As always System Designer is better judge