Session Title:: IBM Power Systems Technical University
Session Title:: IBM Power Systems Technical University
Session Title:
Designing a PowerHA SystemMirror for AIX High Availability Solution
Session ID:
HA17(AIX)
Best Practices for Designing a PowerHA SystemMirror for AIX High Availability Solution
Michael Herrera ([email protected]) Advanced Technical Skills (ATS) Certified IT Specialist
+
Workload-Optimizing Systems
Agenda
Common Misconceptions & Mistakes Infrastructure Considerations Differences in 7.1 Virtualization & PowerHA SystemMirror Licensing Scenarios Cluster Management & Testing Summary
Packaging Changes:
Standard Edition - Local Availability Enterprise Edition - Local & Disaster Recovery
Licensing Changes:
Small, Medium, Large Server Class
Product Lifecycle: Version HACMP 5.4.1 PowerHA 5.5.0 PowerHA SystemMirror 6.1.0 PowerHA SystemMirror 7.1.0 Release Date Nov 6, 2007 Nov 14, 2008 Oct 20, 2009 Sept 10, 2010 End of Support Date Sept, 2011 N/A N/A N/A
4
7.1.0.1 Sep
6.1.0.2 May 21
5.5.0.6 June 7
5.4.1.8 May 13
Common Misconceptions
PowerHA SystemMirror is an out of the box solution
Scripting & Testing of application Start / Stop scripts Application monitors will also require scripting & testing
Fact: Clustering will highlight what you are & are NOT doing right in your environment
6
Lack of Education / Experience Not knowing Expected fallover behaviors Lack of application monitoring Not knowing what to monitor or check
CLI Logs
I/O Pacing Enabled (old values) HBA Levels at GA code Fiber Channel Tunable settings not enabled Interim Fixes not loaded on all cluster nodes
LAN Infrastructure
Redundant Switches
SAN Infrastructure
Redundant Fabric
Application Availability
Application Monitoring Availability Reports
Infrastructure Considerations
Site A
All links through one pipe
Site B
LAN
LAN
SAN
DWDM Node A 50GB 50GB
SITEAMETROVG
SAN
DWDM Node B
50GB 50GB
Important:
Identify & Eliminate Single Points of Failure! 9
Infrastructure Considerations
Site A
XD_rs232 XD_IP
Site B
WAN
net_ether_0
LAN
LAN
SAN
DWDM Node A
ECM VG: diskhb_vg1 hdisk2 000fe4111f25a1d1
SAN
DWDM Node B
ECM VG: diskhb_vg1 hdisk3 000fe4111f25a1d1
1GB
1GB
50GB 50GB
SITEAMETROVG
50GB 50GB
Important:
Identify Single Points of Failure & design the solution around them 10
Infrastructure Considerations
Power Redundancy
Real Customer Scenarios:
I/O Drawers SCSI Backplane SAN HBAs Virtualized Environments Application Fallover Protection
Moral of the Story: * High Availability goes beyond just installing the cluster software
11
diskhb_net1
LPAR 1
diskhb_net2
LPAR 1
LPAR 4
LPAR 2
LPAR 4
LPAR 2
Multicasting
diskhb_net4
LPAR 3
diskhb_net3
LPAR 3
12
en0
en0
(persistent IP) 9.19.51.11 ( base address) 192.168.100.2 HB Rings In 6.1 & below
VLAN
en1 en1
Traditional heartbeating rules no longer apply. However, route stripping is still a potential issue. When two interfaces have routable IPs on the same subnet AIX will send half the traffic out of either interface
Methods to circumvent this: Link Aggregation / EtherChannel Virtualized Interfaces with dual VIO servers
9.19.51.21 9.19.51.20 9.19.51.10 (service IP) (service IP) (base address)
en2
en2
( base address)
9.19.51.11
VLAN
ent0 ent1 ent0 ent1
13
en3
en3
en2
en2
( base address)
9.19.51.11
VLAN
ent0 ent1 ent0 ent1
14
What is it: A set of services/tools embedded in AIX to help manage a cluster of AIX nodes and/or help run cluster software on AIX IBM cluster products (including RSCT, PowerHA, and the VIOS) will use and/or call CAA services/tools CAA services can assist in the management and monitoring of an arbitrary set of nodes and/or running a third-party cluster CAA does not form a cluster by itself. It is a tool set. There is no notion of quorum (If 20 nodes of a 21 node cluster are down, CAA still runs on the remaining node) CAA does not eject nodes from a cluster. CAA provides tools to fence a node but never fences a node and will continue to run on a fenced node
Major Benefits: Enhanced Health Management (Integrated Health Monitoring) Cluster Wide Device Naming
15
RSCT Consumers
IBM Storage HPC PowerHA System Mirror VIOS
Legacy RSCT
Bundled Resource Managers Group Services Messaging API Cluster Messaging Resource Mgr Services Monitoring API Cluster Monitoring Cluster Admin UI Cluster CFG Repository
Cluster Layers Integrated Cluster to CAA Cluster CFG Redesigned Capabilities Messaging Monitoring Repository
RSCT and Cluster Aware AIX together provide the foundation of strategic Power Systems SW RSCT-CAA integration enables compatibility with a diverse set of dependent IBM products RSCT integration with CAA extends simplified cluster management along with optimized and robust cluster monitoring, failure detection, and recovery to RSCT exploiters on Power / AIX 16
Host 1
Host 2
Host 3
Direction: In the first release, support is confined to shared storage Will eventually evolve into a general AIX device rename interface Future direction is to enable clusterwide storage policy settings PowerHA ODM will eventually also entirely move to the repository disk
PowerHA SystemMirror 6.1 & Prior:
Cluster Synchronization
HA ODM
HA ODM
HA ODM
Central Repository
PowerHA SystemMirror will continue to run if Central Repository Disk goes away However, no changes may take place within the cluster.
Host 1
Host 2
Host 3
17
LPAR 1
LPAR 2
Heartbeats
Reliable Messaging
Heartbeats
Reliable Messaging
Network SAN
Repository Disk
Highlights: RSCT Topology services no longer used for cluster Heartbeating All customers now have multiple communication paths by default
18
SAN Network
Site B
SAN Network
Disk Replication
Crossover Connections Virtualized Resources Multiple Resource Groups Mutual Takeover Custom Resource Groups Adaptive Fallover NFS Cross-Mounts File Collections Dependencies Parent / Child Location Start After
Stop After
Smart Assistants Multiple Sites Cross Site LVM Configs Storage Replication IP Replication Application Monitoring Pager Events DLPAR Integration
Grow LPAR on Fallover
One to one
One to any
Any to one
Any to any
20
RG Dependency
Mutual Takeover
RG Dependency
Virtualization
NIC HBA
Frame 2 Node B
hdisk1
SAN
oracle_vg1
{ {
hdisk2
SAN
VIO
NIC
VIO
NIC
hdisk4
VIO
VIO
hdisk4
} }
rootvg
Storage Subsystem
HBA
HBA
oracle_vg1
21
Power HA Cluster
LPAR / DLPAR Micropartitioning & Shared Processor Pools Virtual I/O Server
Virtual Ethernet Virtual SCSI Virtual Fiber
Power HA_node 1
LPAR X
AIX Rootvg
en0 vfc0
LPAR Y
LPAR Z
Data
Data
vfc1
vfc1
LAN
SAN
Storage Virtualization
Both methods of virtualizing storage are supported
VSCSI vs. Virtual fiber (NPIV)
In DR implementations leveraging disk replication consider the implications of using either option
Benefits of virtualization:
Maximize utilization of resources Less PCI slots & physical adapters Foundation for advanced functions like Live Partition Mobility Migrations to newer Power Hardware are simplified
* Live Partition Mobility & PowerHA SM compliment each other Maintenance vs. High Availability
(non-reactive . reactive)
23
PowerHA LPAR 1
en0
PowerHA LPAR 2
en0
ent4 (SEA)
ent0 (phy)
ent2 (virt)
ent5 (virt)
ent6 (virt)
ent0 (virt)
ent0 (virt)
ent6 (virt)
ent5 (virt)
ent2 (virt)
ent0 (phy)
PVID 99
Hypervisor Frame 1
PVID 10
Ethernet Switch
Ethernet Switch
This is a diagram of the configuration required for SEA fallover across VIO Servers. Note that Ethernet traffic will not be load balanced across the VIO Servers. The lower trunk priority on the ent2 virtual adapter would designate the primary VIO Server to use.
24
PowerHA LPAR 1
en0
Control Channel
Control Channel
Frame1
ent1 (phy)
ent0 (phy)
ent2 (virt)
ent5 (virt)
ent0 (virt)
ent5 (virt)
ent2 (virt)
ent1 (phy)
ent0 (phy)
Hypervisor
Ethernet Switch
Ethernet Switch
Hypervisor
Frame2
ent1 (phy)
ent0 (phy)
ent2 (virt)
ent5 (virt)
Control Channel
ent0 (virt)
ent5 (virt)
Control Channel
ent2 (virt)
ent1 (phy)
ent0 (phy)
ent3 (LA)
ent4 (SEA)
en0
ent4 (SEA)
ent3 (LA)
PowerHA LPAR 2
25
Topsvcs heartbeating
en0
serial_net_0
en0
Hypervisor
ent1 (phy)
ent0 (phy)
ent2 (virt)
ent5 (virt)
Control Channel
FRAME X
ent3 (LA) ent4 (SEA)
ent0 (virt)
ent5 (virt)
Control Channel
ent2 (virt)
ent1 (phy)
ent0 (phy)
en0
ent4 (SEA)
ent3 (LA)
26
Differences: RSCT Topology services is no longer used for heartbeat monitoring Subnet Requirements no longer need to be followed Netmon.cf file is no longer required or used All interfaces are used for monitoring even if they are not in an HA network (this may be tunable in a future release) IGMP Snooping must be enabled on the switches
27
Node 1 vscsi0
hdisk0
} rootvg
HBA HBA
vhost0
Hypervisor
MPIO
vscsi1
hdisk1 hdisk2
hdisk
fcs0
MPIO
fcs1
hdisk3 hdisk4
VIOS 2
} }
vscsi_vg
npiv_vg
FRAME 2
STORAGE SUBSYSTEM
VIOS 1
Node 2
hdisk0
} rootvg
HBA HBA
vhost0
MPIO
vscsi1 fcs0
hdisk1 hdisk2
} }
vscsi_vg
hdisk
MPIO
fcs1
hdisk3 hdisk4
npiv_vg 28
VIOS 2
Frame 1
Frame 2
Considerations: This is a planned move It assumes that all resources are virtualized through VIO (Storage & Ethernet connections) PowerHA should only experience a minor disruption to the heartbeats during a move IVE / HEA virtual Ethernet is not supported for LPM VSCSI & NPIV virtual fiber mappings are supported
VIOS 1 VIOS 2
VIOS 1 VIOS 2
rootvg
datavg
The two solutions compliment each other by providing the ability to perform non-disruptive maintenance while retaining the ability to fallover in the event of a system or application outage 29
*~ 2 seconds of total interruption time ** Require free system resources on target system
30
Results:
120GB DLPAR add took 1min 55 sec 246GB DLPAR add took 4 min 25 sec 30% busy running artificial load the add took 4 minutes 36 seconds
ssh communication
ssh communication
LPAR A HMC
DLPAR CPU Count Minimal CPU Count Application Server
LPAR B HMC
Minimal CPU Count
Backup
31
System A
Acquired via DLPAR with App + 1 CPU + 2 CPU Oracle DB 1 CPU Banner DB 1 CPU Standby Standby 1 CPU 1 CPU Cluster 1 Cluster 2 Cluster 3 Cluster 4
System B
Standby Standby 1 CPU 1 CPU + 1 CPU Acquired via DLPAR + 2 CPU with App
Applications
Production Oracle DB Production PeopleSoft AIX Print Server Banner Financial DB Production Financial DB Tivoli Storage Manager 5.5.2.0
CPU
2 2 2 3 3 2
Memory
16 GB 8 GB 4 GB 32 GB 32 GB 8 GB
32
System A
Oracle DB 1 CPU Banner DB 1 CPU Standby Standby 1 CPU 1 CPU Cluster 1 Cluster 2 Cluster 3 Cluster 4
System B
Standby Standby 1 CPU 1 CPU
The actual application requirements are stored in the PowerHA SystemMirror definitions and enforced during the acquisition or release of application server resources
HMC
System A
Acquired via DLPAR with App + 1 CPU + 2 CPU Oracle DB 1 CPU Banner DB 1 CPU Standby Standby 1 CPU 1 CPU Cluster 1 Cluster 2 Cluster 3 Cluster 4
System B
Standby Standby PeopleSoft 1 CPU 1 CPU 1 CPU + 1 CPU
During acquisition of resources in the cluster start up the host will ssh to the pre-defined HMC/s to perform the DLPAR operation automatically
Financial DB 1 CPU
33
2. Start PowerHA
Application Server Min 1 Desired 2 Max 2 Application Server Min 1 Desired 3 Max 3
Read Requirements
Application Server Min 1 Desired 2 Max 2 Application Server Min 1 Desired 3 Max 3
Activate LPARs
LPAR Profile Min 1 Desired 1 Max 2 LPAR Profile Min 1 Desired 1 Max 3
HMC
DLPAR
DLPAR
System A
- 1 CPU - 2 CPU 3. Release resources Fallover or RG_move + 1 CPU + 2 CPU 1 CPU Oracle DB 2 1 CPU Banner DB 3 Cluster 1 Cluster 2
System B
Standby Oracle DB 12CPU CPU Standby CPU Banner DB 1 3 CPU + 1 CPU + 2 CPU - 1 CPU - 2 CPU 4. Release resources Stop cluster without takeover
Take Aways: CPU allocations follow the application server wherever it is being hosted (this model allows you to lower the HA license count) DLPAR resources will only get processed during the acquisition or release of cluster resources PowerHA 6.1+ allows provide micro-partitioning support and the ability to also alter virtual processor counts DLPAR resources can come from free CPUs in shared processor pool or CoD resources
34
System B
Standby Standby 2 CPU 3 CPU
PowerHA license counts: Cluster 1 : 4 CPUs Cluster 2 : 6 CPUs Cluster 3 : 4 CPUs Cluster 4 : 6 CPUs Total : 20 licenses
HMC
System B
Standby Standby PeopleSoft 1 CPU 1 CPU 1 CPU
PowerHA license counts: Cluster 1 : 3 CPUs Cluster 2 : 4 CPUs Cluster 3 : 3 CPUs Cluster 4 : 4 CPUs Total : 14 licenses
+ 1 CPU
Financial DB 1 CPU
35
+ 1 CPU + 2 CPU
PowerHA license counts: Cluster 1 : 3 CPUs Cluster 2 : 4 CPUs Cluster 3 : 3 CPUs Cluster 4 : 4 CPUs Total : 14 licenses
+ 1 CPU
Financial DB 1 CPU
HMC
System B
36
* Consolidated both Prod LPARs into one LPAR. Control separated by Resource Groups
37
Node B
Datavg
ECM VG
Cluster Services
Cluster Services
PASSIVE
read only
Shared LUNs 38
IBM Systems Director Plug-in New for PowerHA SystemMirror 7.1 Only for management of 7.1 & above Same look and feel as IBM suite of products Will leverage existing Director implementation Uses clvt & clmgr CLI behind the covers
39
User_1
User_2
User_3
User_4
Multiple WebSMIT users accessing multiple clusters through *one* WebSMIT server
Cluster_A
Cluster_C
40
41
Three tier architecture provides scalability: User Interface Management Server Director Agent
Director Agent
Automatically installed on AIX 7.1 & AIX V6.1 TL06
P P P P
D D D
Secure communication D
Director Server
P P P
Central point of control Supported on AIX, Linux, and Windows Agent manager
42
43
44
45
46
47
48
From this release forward, only clmgr is supported for customer use
clvt is strictly for use by the Smart Assists
bos.cluster.rte
# clcmd lssrc -g caa ------------------------------NODE mutiny.dfw.ibm.com ------------------------------Subsystem Group PID clcomd caa 9502848 cld caa 10551448 clconfd caa 10092716 solid caa 7143642 solidhac caa 7340248 ------------------------------NODE munited.dfw.ibm.com ------------------------------Subsystem Group PID cld caa 4390916 clcomd caa 4587668 clconfd caa 6357196 solidhac caa 6094862 solid caa 6553698
# clcmd lspv ------------------------------NODE mutiny.dfw.ibm.com ------------------------------hdisk0 0004a99c161a7e45 caa_private0 0004a99cd90dba78 hdisk2 0004a99c3b06bf99 hdisk3 0004a99c3b076c86 hdisk4 0004a99c3b076ce3 hdisk5 0004a99c3b076d2d ------------------------------NODE munited.dfw.ibm.com ------------------------------hdisk0 0004a99c15ecf25d caa_private0 0004a99cd90dba78 hdisk2 0004a99c3b06bf99 hdisk3 0004a99c3b076c86 hdisk4 0004a99c3b076ce3 hdisk5 0004a99c3b076d2d
50
Attention: Sendmail must be working and accessible via the firewall to receive notifications
52
Syncd Setting
Default value of 60 recommended change to 10
Failure Detection Rate (FDR) only for Version 6.1 & below
Normal Settings should suffice in most environments (note that it can be tuned further) Remember to enable FFD when using disk heartbeating
53
Test Cluster
Lpars within same frame Virtual resources
Utilize Available Tools Cluster Test Tool Testing upgrades Alternate disk install is your friend
Best Practice: Testing should be the foundation for your documentation in the event that someone not PowerHA savvy is there when a failure occurs.
54
munited /# cltopinfo -m Interface Name Adapter Total Missed Current Missed Address Heartbeats Heartbeats -------------------------------------------------------------------------------------------------------------------en0 192.168.1.103 0 0 rhdisk1 255.255.10.0 1 1 Cluster Services Uptime: 30 days 0 hours 31 minutes
55
Summary
Review your infrastructure for potential single points of failure
Be aware of the potential pitfalls listed in the common mistakes slide
Popular Topics: * Frequently Asked Questions * Customer References * Documentation * White Papers
http://www-03.ibm.com/systems/power/software/availability/aix/index.html
( or Google PowerHA SystemMirror and click Im Feeling Lucky)
57
Questions?
Additional Resources
New - Disaster Recovery Redbook
SG24-7841 - Exploiting PowerHA SystemMirror Enterprise Edition for AIX
http://www.redbooks.ibm.com/abstracts/sg247841.html?Open
Online Documentation
http://www-03.ibm.com/systems/p/library/hacmp_docs.html
59