Professional Documents
Culture Documents
HP Poli Serv
HP Poli Serv
00
Use of this material to deliver training without prior written permission from HP is prohibited.
hf836s c.00
Use of this material to deliver training without prior written permission from HP is prohibited.
Module 10 Troubleshooting
Network and SAN issues ............................................................................ 10-3
Membership partitions .............................................................................. 10-20
Other troubleshooting issues .......................................................................10-25
Lab activity.............................................................................................. 10-53
hf836s c.00
Course Introduction
Module 0
hf836s c.00
HP Education Services
2009 Hewlett-Packard Development Company, L.P.
The information contained herein is subject to change without notice
hf836s c.00
0-1
Course Introduction
Course objectives
At the end of this course, you will have:
A thorough understanding of the HP PolyServe Linux product
Enough hands-on experience to enable the successful
0-2
hf836s c.00
0-2
Course Introduction
Course prerequisites
If you are taking this course, you should have:
A rudimentary knowledge of clustering technologies
A working knowledge of SAN technologies
An in-depth knowledge of LinuxOS administration
A working knowledge of Apache, Oracle, NFS
0-3
hf836s c.00
0-3
Course Introduction
Agenda
Day One
Module 1 Product Overview
Module 2 Installation and Configuration
Module 3 Storage Configuration
0-4
Agenda
The time spent on each module will vary depending on the audience.
hf836s c.00
0-4
Course Introduction
Agenda
Day Two
Module 4 File Systems and Quotas
Module 5 Matrix Administration
Module 6 Matrix Server Clusterware
Module 7 File Serving Utility
0-5
Agenda
The time spent on each module will vary depending on the audience.
hf836s c.00
0-5
Course Introduction
Agenda
Day Three
Module 7 File Serving Utility (contd)
Module 8 Database Utility
Module 9 Logs and Services
Module 10 Troubleshooting
0-6
Agenda
The time spent on each module will vary depending on the audience.
hf836s c.00
0-6
Course Introduction
Classroom facilities
Fire exits
Phones
Restrooms
Smoking
Breaks
Lunch
Class time
Begin class promptly
Use time wisely
Use the student guide as a reference manual
No cell phones or laptops please
0-7
hf836s c.00
0-7
Course Introduction
Introductions
Name
Work location
Years with company
Experience
Certifications
Expectations for this course
0-8
hf836s c.00
0-8
hf836s c.00
Course Introduction
2009 Hewlett-Packard
Rev. 7.4 Development
HP Restricted Company, L.P.
0-9
hf836s c.00
0 - 10
Course Introduction
MxS Overview
Module 1
hf836s c.00
hf836s c.00
1-1
MxS Overview
Objectives
At the end of this module the student should be able to:
Describe HP PolyServe Software and its components used in the
Scalable NAS Products
Describe the type of environments in which the HP PolyServe
Software can be deployed
Understand hardware requirements for servers running HP
PolyServe Software
Explain HP PolyServe clustering
1-2
hf836s c.00
hf836s c.00
1-2
MxS Overview
Cluster-wide administration
Management Console and command line interface enable
configuration and management of the entire matrix remotely or from
any server in the matrix
1-3
hf836s c.00
hf836s c.00
1-3
MxS Overview
Fully symmetric
High-Availability Services
MPIO
Dual HBA
Dual Switch
Dual Fabric
FC SW
RAID
Array
FC SW
RAID
Array
hf836s c.00
hf836s c.00
1-4
MxS Overview
Cluster overview
Public
Public
Network(s)
Network(s)
Server 1
Server 2
Server 3
Fibre Channel/ISCSI
The structure of
Ethernet/IP
100/1000/10000
1-5
hf836s c.00
a 4-node cluster
hf836s c.00
1-5
Server 4
MxS Overview
clienty
clientz
client
client
client
client
NAS
Head
client
clienty
clientz
NAS
Head
NAS
Head
NAS
Head
client
NAS
Head
NAS
Head
clientx
clientx
clienty
clientz
1-6
hf836s c.00
hf836s c.00
1-6
clienty
clientz
HP PolyServe features
MxS Overview
(1 of 4)
Storage Features
Shared File System
General purpose cluster file system
Fully symmetric, concurrent access by multiple servers in the
matrix
Any server can lock a file
Any server can update metadata
1-7
hf836s c.00
hf836s c.00
1-7
HP PolyServe features
MxS Overview
(2 of 4)
1-8
hf836s c.00
hf836s c.00
1-8
HP PolyServe features
MxS Overview
(3 of 4)
Service/Application monitors
A network service or application such as HTTP, NFS or Oracle
Monitors the application and will cause the vhost to failover in the
event of a failure
Device monitors
Designed to watch a part of a server, such as a shared file system,
and failover or influence the location of a virtual host
File-based Replication
Cluster-aware replication for disaster recovery purposes
1-9
hf836s c.00
hf836s c.00
1-9
HP PolyServe features
MxS Overview
(4 of 4)
Primary Solutions
Web Serving
Deploy a highly available scale-out web site using Industry
Standard Servers and Storage
File Serving
Deploy a highly available scale-out NAS utility using Industry
Standard Servers and Storage
1 - 10
hf836s c.00
hf836s c.00
1 - 10
MxS Overview
HP PolyServe processes
User Mode
Software Components
Kernel Mode
ClusterPulse
PANPulse
GroupCom
Management
Console
HBA
driver
SDMP
psd
driver
SCL
PSFS
SANPulse
DLM
LCL
pswebsrv
mxinit
mxlogd
1 - 11
hf836s c.00
hf836s c.00
1 - 11
mxlog
MxS Overview
OS:
Storage:
SAN storage devices. The Clustered File System is designed to be scalable, highly
recoverable and highly available
cluster
manage the entire Cluster either remotely or from any server in the Cluster.
1 - 12
hf836s c.00
hf836s c.00
1 - 12
MxS Overview
1 - 13
hf836s c.00
hf836s c.00
1 - 13
MxS Overview
HP StorageWorks ExDS9100
System Specifications
System
Power*
Weight
Performance Capability
Thermal
1 - 14
hf836s c.00
hf836s c.00
1 - 14
MxS Overview
HP ExDS Software
Linux OS
Red Hat Enterprise Linux v4 Update 4
HP PolyServe HA Services
Full n:1, n-m monitoring and failover for HTTP
and NFS v3
1 - 15
hf836s c.00
hf836s c.00
1 - 15
MxS Overview
128 vhosts
128 server and/or device monitors
4 HBA Ports per server
1 - 16
hf836s c.00
hf836s c.00
1 - 16
hf836s c.00
1 - 17
MxS Overview
MxS Overview
hf836s c.00
Check the online Compatibility Matrix for an up to date list of supported hardware and
software:
hf836s c.00
1 - 18
MxS Overview
Local Storage
105MB disk space for the MxS software
100MB disk space for log and runtime files
1 - 19
hf836s c.00
Memory resources on each server in the matrix are consumed to manage the state that
preserves the coherency of shared file systems.
For this reason, all servers should have approximately the same memory footprint. A 2-to-1
ratio typically is not problematic, but a larger difference can increase paging and affect lock
management to the extent that overall matrix performance is impacted.
Failover Considerations
It is important to understand your failover requirements such that any server that is a
candidate for failover has the required hardware to support that application.
hf836s c.00
1 - 19
MxS Overview
Cluster design
Time Synchronization
To ensure that file times are consistent across the matrix, it is
important that all servers operate with synchronized time-ofday clocks. An NTP server is one commonly used mechanism
for synchronizing system clocks.
Security
When configuring security on PSFS file systems, you need to
ensure the uids and gids are the same on every node.
1 - 20
hf836s c.00
hf836s c.00
1 - 20
MxS Overview
Consider using the /etc/hosts file for name resolution of nodes in the
matrix rather than an external DNS server
Administrative Network
1 - 21
hf836s c.00
hf836s c.00
1 - 21
MxS Overview
This is how users get access to the applications running in the cluster. It
is a best practice to keep the public and private networks separate
1 - 22
hf836s c.00
hf836s c.00
1 - 22
MxS Overview
hf836s c.00
hf836s c.00
1 - 23
MxS Overview
10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Half 1000baseT/Full
Supports auto-negotiation: Yes
Advertised link modes: 10baseT/Half 10baseT/Full
100baseT/Half 100baseT/Full
1000baseT/Half 1000baseT/Full
Advertised auto-negotiation: Yes
Speed: 1000Mb/s
Duplex: Full
Port: Twisted Pair
PHYAD: 1
Transceiver: internal
Auto-negotiation: on
Supports Wake-on: g
Wake-on: d
Current message level: 0x000000ff (255)
Link detected: yes
1 - 24
hf836s c.00
hf836s c.00
1 - 24
MxS Overview
1 - 25
hf836s c.00
hf836s c.00
1 - 25
MxS Overview
1 - 26
hf836s c.00
hf836s c.00
1 - 26
MxS Overview
Storage
Generally speaking HP PolyServe works with all storage
The main issue is to understand any multipath requirements of the array
and the deployment
Switch attached storage is recommended (presents WWNN:LUN#).
In some cases, scsi code page information is used to derive a unique ID
1 - 27
hf836s c.00
hf836s c.00
1 - 27
MxS Overview
hf836s c.00
hf836s c.00
1 - 28
MxS Overview
hf836s c.00
Fabric based fencing is preferable and approximately 90% of our end users use it.
hf836s c.00
1 - 29
MxS Overview
Pros:
Cons
Requires out of band snmp access to FC switches
1 - 30
hf836s c.00
Fabric based fencing is preferable and approximately 90% of our end users use it.
hf836s c.00
1 - 30
MxS Overview
Pros
Does not require FC switches or access to FC switches
Cons
Cannot take a crash dump on a node when using NFF
Requires operator input to setup
1 - 31
hf836s c.00
Fabric based fencing is preferable and approximately 90% of our end users use it.
hf836s c.00
1 - 31
MxS Overview
Ready to Install
Once the Servers are setup correctly with
A supported Operating System
A good network configuration
A good storage configuration
Suitable HBA drivers, if the drivers delivered with the PolyServe
software are not being used
Appropriate MPIO s/w
then we are ready to install the cluster software.
1 - 32
hf836s c.00
hf836s c.00
1 - 32
hf836s c.00
1 - 33
MxS Overview
hf836s c.00
1 - 34
MxS Overview
Module 22
c.00
hf836s c.00
hf836s c.00
2-1
Objectives
At the end of this module the you should be able to:
Install HP PolyServe for Linux
Configure a HP PolyServe Cluster
2-2
hf836s c.00
hf836s c.00
2-2
hf836s c.00
2-3
Prerequisites
If the HP provided kernel patches are required then a kernel rebuild may
need to be performed but a binary kernel precompiled with any patches
is available from HP
2-4
hf836s c.00
lmsensors, netsnmp,
These are open source kernel patches that HP required to complete QA of the
product
If the File Serving solution pack is being used then a kernel rebuild or HP
binary kernel is required for the NFS changes
2-4
of the networks
2-5
hf836s c.00
hf836s c.00
2-5
Ensure broadcast and multicast are enabled on the network interface and
switches
Unicast can be supported via a configuration parameter
hf836s c.00
hf836s c.00
2-6
2-7
hf836s c.00
Linux
configuration
option
Default
Value
Recommended
Value
ethtool k eth0
on
off
Number of receive
Varies by NIC
(Rx) and transmit (Tx) ethtool g eth0
hardware
Buffers (FIFOs)
Balanced
maximums,
favoring receive
ethtool a eth0 on
off
ifconfig eth0
30000 1GbE
30000 10GbE
NIC Statistics
1000
hf836s c.00
2-7
ethtool -a eth0
ethtool -a eth0
tx off
rx off
2-8
hf836s c.00
hf836s c.00
2-8
2-9
hf836s c.00
hf836s c.00
2-9
NIC Bonding
balance-rr or 0 - Round-robin policy: Transmit packets in sequential
order from the first available slave through the last. This mode provides
load balancing and fault tolerance.
active-backup or 1 - Active-backup policy: Only one slave in the
bond is active. A different slave becomes active if, and only if, the
active slave fails. The bond's MAC address is externally visible on only
one port (network adapter) to avoid confusing the switch.
balance-xor or 2 - XOR policy: Transmit based on the selected
transmit hash policy. The default policy is a simple [(source MAC
address XOR'd with destination MAC address) modulo slave count].
Alternate transmit policies may be selected via the xmit_hash_policy
option.
2 - 10
hf836s c.00
hf836s c.00
2 - 10
2 - 11
hf836s c.00
hf836s c.00
2 - 11
/etc/modprobe.conf.local
alias bond0 bonding
options bonding miimon=500 mode=802.3ad lacp_rate=0
use_carrier=1 xmit_hash_policy=layer3+4
Reference documents
2 - 12
hf836s c.00
/usr/src/linux/Documentation/networking/bonding.txt/
Novell search for Setting Up Bonding on SLES 9
hf836s c.00
2 - 12
2 - 13
hf836s c.00
hf836s c.00
2 - 13
2 - 14
hf836s c.00
hf836s c.00
2 - 14
HBA Drivers
HBA drivers from Qlogic and Emulex are supported, HP do not develop
modprobe.conf
2 - 15
hf836s c.00
hf836s c.00
2 - 15
Note: Partition tables are NOT required on luns being used for file
systems but they ARE required for the membership partition luns
2 - 16
hf836s c.00
hf836s c.00
2 - 16
ql2xlbType=0
(not used when ql2xfailover set to 0 and not required
as we are not using the f/o driver in v3.7)
ql2xmaxqdepth=32
qlport_down_retry=30
ql2xloginretrycount=30
ql2xexcludemodel=0x0
ql2xautorestore=0xA6
2 - 17
hf836s c.00
hf836s c.00
2 - 17
# cat /etc/hba.conf
qla2xxx /opt/polyserve/lib/hba_files/qla2xxx-8.02.11/libqlsdm.so
2 - 18
hf836s c.00
hf836s c.00
2 - 18
hf836s c.00
hf836s c.00
2 - 19
2 - 20
hf836s c.00
hf836s c.00
2 - 20
Cluster.
Partition disks before enabling DM_MPIO or a reboot
may be required to recognize new partitions.
Refer to HP Scalable NAS Documentation
Refer to OS Distribution Documentation.
http://www.redhat.com/docs/manuals/enterprise/RHEL-5manual/en-US/RHEL510/DM_Multipath/index.html
http://www.novell.com/documentation/sles10/stor_evms/index.ht
ml?page=/documentation/sles10/stor_evms/data/multipathing.htm
l
2 - 21
hf836s c.00
The cluster software expects that if the device mapper multipath is to be used, that it is
installed and configured before starting the cluster. Our install and upgrade documents talk
about how to do this. Also, there is documentation from the OS distribution that tells how to
set up device mapper multipath.
Our documentation only mentions using the HP device mapper enablement package to get
the required multipath.conf file and setting for HBA parameters. The correct multipath.conf
settings for other vendor's storage would have to come from the vendor.
MP disk partitions should be created before starting device mapper multipath. Device mapper
and partitions don't work particularly well with each other. If a disk has partitions already on
it, that is fine. If you want to put partitions on a disk that is already controlled by dm-mpio,
then it is more trouble. The distribution documentation says to reboot after creating partitions.
This is because re-reading the partition data when the device is controlled by the device
mapper doesn't work. This is why, to make things easier, partitions should be added to MP
disks before enabling device mapper multipath.
hf836s c.00
2 - 21
Download the HPDM Multipath Enablement Kit for HP StorageWorks Disk Arrays
v4.2.0 available
at http://www.hp.com/go/devicemapper.
#cd /tmp/HPDMmultipath
#tar -xvzf HPDMmultipath-4.2.0.tar.gz
#cd HPDMmultipath-4.2.0
Verify that the directory contains README.txt, COPYING, INSTALL, bin, conf,
SRPMS, and docs directories
2 - 22
hf836s c.00
hf836s c.00
2 - 22
hf836s c.00
hf836s c.00
2 - 23
*********************************************************************
Installation completed successfully!
[root@poly2 HPDMmultipath-4.2.0]#
2 - 24
hf836s c.00
hf836s c.00
2 - 24
Path grouping policies: Paths are coalesced based on the following pathgrouping policies:
Priority based path-grouping
Provides priority to group paths based on Asymmetric Logical Unit Access (ALUA) state
Provides static load balancing policy by assigning higher priority to the preferred path
Multibus
All paths are grouped under a single path group
Group by serial
Paths are grouped together based on controller serial number
Failover only
Provides failover without load balancing by grouping the paths into individual path groups
I/O load balancing policies: Provides weighted Round Robin load balancing policy within a path group
I/O load balancing policies: Provides weighted Round Robin load balancing
policy within a path group
Path monitoring: Periodically monitors each path for status and enables faster
failover and failback
2 - 25
hf836s c.00
hf836s c.00
2 - 25
"HP|COMPAQ"
product
"HSV1[01]1 \(C\)COMPAQ|HSV[2][01]0|HSV300"
path_grouping_policy
group_by_prio
getuid_callout
"/sbin/scsi_id -g -u -s /block/%n"
path_checker
tur
path_selector
"round-robin 0"
prio_callout
"/sbin/mpath_prio_alua /dev/%n"
rr_weight
uniform
failback
immediate
hardware_handler
no_path_retry
rr_min_io
"0"
12
100
2 - 26
hf836s c.00
hf836s c.00
2 - 26
multipath.conf entry - XP
For XP
device {
vendor
"HP"
product
"OPEN-.*"
path_grouping_policy
multibus
getuid_callout
"/sbin/scsi_id -g -u -s /block/%n"
path_selector
"round-robin 0"
rr_weight
uniform
path_checker
tur
hardware_handler
"0"
failback
immediate
no_path_retry
12
rr_min_io
1000
2 - 27
hf836s c.00
hf836s c.00
2 - 27
2 - 28
hf836s c.00
Partitions should be created when the device is not part of a device mapper multipath device.
If you run fdisk on a dm- device or one of its component paths, you will see this when you try
to write the partition table out:
WARNING: Re-reading the partition table failed with error 22: Invalid argument.
The kernel still uses the old table.
The new table will be used at the next reboot.
Syncing disks.
The versions of the multipath tools are different between SLES10 and RHEL5. The kpartx
program functions a little differently between the two. Kpartx is used by udev to create linear
devices mapped to the multipath device to act as the partition devices. The RHEL5 version
does not generate a map for the extended partition. The SLES10 version of kpartx does.
hf836s c.00
2 - 28
2 - 29
hf836s c.00
multipathd logs messages to /var/log/messages. You can see when paths fail over or fail
back.
sandiskinfo shows what device mapper devices and which sd devices it is managing for each
psd.
dmsetup is a device mapper utility managing mapped devices. It can also be used to list
mapped devices, remove mapped devices, and display information about mapped devices.
"dmsetup ls --tree" will show a tree of multipath devices and their component sd devices.
"dmsetup remove_all" will remove all mapped devices (if not is use) force will remove in use
mapped devices as well.
The multipath command is used to create mapped multipath devices. It can also be used to
list the current devices. Use "multipath -ll" it produces something like "dmsetup ls --tree" but
with more information.
"multipathd -k" will start a copy of the multipathd in interactive mode and you can display
multipathd state. Use control D to exit.
hf836s c.00
2 - 29
hf836s c.00
hf836s c.00
2 - 30
Failover Timeout
psd1
disabled
psd1p1
disabled 10000
psd2
disabled
psd2p1
psd3
disabled 10000
disabled
none
psd3p1
disabled 10000
psd4
disabled
none
psd5
disabled
none
psd6
disabled
none
psd7
none
2 - 31
hf836s c.00
disabled
none
Targets
none
hf836s c.00
2 - 31
Server
This is run automatically when the product is installed and every time the product is
started
It can also be run interactively at any time
hf836s c.00
hf836s c.00
2 - 32
2 - 33
hf836s c.00
hf836s c.00
2 - 33
1.
2.
rpm i pmxs-X.Y.Z-<xxxx>.<arch>.rpm
rpm i pmxs-<os>-support.X.Y.Z-<xxx>.<arch>.rpm
rpm I kernel-HPPS-<kernel_version>.<xxxx>.<arch>.rpm
or
3.
4.
5.
rpm i mxconsole-X.Y.Z-<xxxx>.<arch>.rpm
2 - 34
hf836s c.00
hf836s c.00
2 - 34
2 - 35
hf836s c.00
hf836s c.00
2 - 35
2 - 36
hf836s c.00
hf836s c.00
2 - 36
2 - 37
hf836s c.00
hf836s c.00
2 - 37
iLO Port
Private Intra-Cluster Network
PCI-X PCI-E
3
100
MHz
2
x4
2
100
MHz
1
x4
iLO
1
133
MHz
N/A
SCSI Port 1
UID
HBA Ports
Embedded NIC Port 2 Public
Data (NFS) Network
2 - 38
hf836s c.00
hf836s c.00
2 - 38
3.
Assign the same user/password to each node for fencing (can use
the admin user if desired).
Set network params (IP, Mask, etc) for private network fencing
Display Setup
2 - 39
hf836s c.00
When booting for the first time, you will need to configure the
system for the monitor being used. Be sure to pick a configuration
that allows for 16K colors (8K color depth will make using the
GUIs difficult) and at least 1024x768 resolution.
hf836s c.00
2 - 39
2 - 40
hf836s c.00
hf836s c.00
2 - 40
Configure Networking
2 - 41
hf836s c.00
hf836s c.00
2 - 41
2 - 42
hf836s c.00
hf836s c.00
2 - 42
address of the server. Then type admin for both user and
password and click Configure
Click Here to
Launch
2 - 43
hf836s c.00
hf836s c.00
2 - 43
hf836s c.00
2 - 44
hf836s c.00
hf836s c.00
2 - 45
2 - 46
hf836s c.00
hf836s c.00
2 - 46
username and
password
On v3.7 this will be
a valid Linux
username/password
as the internal
admin user is no
longer supported
2 - 47
hf836s c.00
hf836s c.00
2 - 47
tab:
2 - 48
hf836s c.00
hf836s c.00
2 - 48
Fencing overview
When problems occur on a server and the server ceases to
2 - 49
hf836s c.00
hf836s c.00
2 - 49
2 - 50
hf836s c.00
hf836s c.00
2 - 50
management hardware on
the server to reboot or
power down the node
Check the compatibility
guide for supported
interfaces
Make sure you know the
IP_address/Username/Pa
ssword of each of the
management ports on the
cluster nodes
2 - 51
hf836s c.00
hf836s c.00
2 - 51
(2 of 2)
2 - 52
hf836s c.00
hf836s c.00
2 - 52
Membership partitions
(1 of 2)
access to the SAN and to store the device and server naming
databases including the global device names that Matrix
Server assigns to the SAN disks placed under its control
Membership partitions should exist on their own LUNs to
prevent any I/O contention
Membership Partitions need to be at least 1GB
MPs need to have a Linux partition table installed on them
We recommend configuring 3 membership partitions for
redundancy
2 - 53
hf836s c.00
Membership partitions
LUNs
A LUN that has a membership partition on it can not be deported while the matrix is
running, Matrix Server needs to be stopped to modify the partition table.
LUNs should already be partitioned and unformatted with no drive letter attached.
hf836s c.00
2 - 53
partitions
2 - 54
hf836s c.00
hf836s c.00
2 - 54
Membership partitions
(2 of 2)
1. To create a membership
Add Membership
Partition window lists all
of the disks or LUNs that
it can access.
2 - 55
hf836s c.00
hf836s c.00
2 - 55
Snapshot Configuration
Click Add to configure
the array based
snapshots
2 - 56
hf836s c.00
hf836s c.00
2 - 56
Snapshot Configuration
1. To create configure array
2 - 57
hf836s c.00
hf836s c.00
2 - 57
Click Apply to
save the configuration
2 - 58
hf836s c.00
hf836s c.00
2 - 58
Cluster-wide configuration
Click Add Server to
add more nodes to
the cluster
Highlight the Server
and then Click
Export To
Highlight the Server
and then Click Start
Service
If using NFF then
highlight the
server and Click
on Test Fencing
2 - 59
hf836s c.00
hf836s c.00
2 - 59
2 - 60
hf836s c.00
Open Management
Console
To Connect to Cluster,
enter either the hostname
or IP address of one of
the servers in the cluster
Enter
username/password to
connect to the cluster
HP PolyServe
installation and
configuration is complete
and you have a running
cluster
hf836s c.00
2 - 60
2 - 61
hf836s c.00
It is important to set up a
dedicated administrative
network that will be used
for cluster interconnect
traffic
Discourage Admin
Traffic on all networks
except those to be used
for admin traffic
You can also Exclude
Admin traffic from ever
running on networks,
except for the primary
network, it can never be
excluded
hf836s c.00
2 - 61
SizingActions file
HP Scalable NAS includes a script called SizingActions that configures certain
operating system parameters to improve system performance, particularly in a
file serving environment. The changes improve network throughput and make
better use system memory. On HP Scalable NAS Clustered Gateway servers
and HP 4400 Scalable NAS File Services systems, additional changes are
made to tune the operating system for the hardware provided with those
systems.
The SizingActions script is run when HP Scalable NAS starts up. The script
does not determine whether the system parameters it adjusts have been
modified from their default values by a user on the system. This can be an
issue if, for example, you are running an application that requires system
parameters such as vmem_max or mmem_max to be modified, typically in the
/etc/sysctl.conf file.
To disable SizingActions in the cluster:
1. Go to the directory containing the SizingActions script:
# cd /etc/opt/hpcfs
2. Run the following command:
# chmod 444 SizingActions
2 - 62
hf836s c.00
hf836s c.00
2 - 62
Other procedures
Hosting Preference
2 - 63
hf836s c.00
hf836s c.00
2 - 63
a failed node.
2 - 64
hf836s c.00
hf836s c.00
2 - 64
hf836s c.00
2 - 65
hf836s c.00
2 - 66
Storage Configuration
Module 3
hf836s c.00
hf836s c.00
3-1
Storage Configuration
Objectives
At the end of this module you should be able to:
Describe Node fencing and its configuration
Describe Membership Partitions and Matrix Data Store
Describe cluster-wide Device Naming
Describe the components of the Clustered Volume Manager
Create Matrix volumes
3-2
hf836s c.00
3-2
hf836s c.00
Storage Configuration
3-3
Split-brain resolution
Storage Configuration
(1 of 2)
protect file system integrity, these are stored in the membership partitions
If a problem causes a cluster to split into two or more network partitions,
the SANlocks ensure that only one of the resulting network partitions has
access to the SAN
this avoids the classic split-brain scenario where two sets of servers
assume they own the data
A server becomes the SAN administrator when it acquires the SANlocks
3-4
Split-brain resolution (1 of 2)
All I/O is suspended and servers update their server registry
entries in the membership partition, then an election
mechanism begins for the SAN administrator role.
hf836s c.00
3-4
Split-brain resolution
Storage Configuration
(2 of 2)
All servers throw away any locks they dont own and all
servers calculate the new home nodes for the locks they own
Another server replays the file system journals to complete
any transactions extant at the time of the crash
Takes only a few seconds to complete
rebooted
3-5
Split-brain resolution (2 of 2)
Typical reasons for nodes getting fenced are:
Network isolation at the node level
Network communication issues between nodes
Lack of udp traffic between the nodes
Lack of multicast communication between the nodes
Network addresses are changed dynamically with MxS
running
Nodes are so heavily loaded they are not communicating
correctly
Network traffic is so high that messages are taking too long
to get between the nodes
hf836s c.00
3-5
Storage Configuration
hf836s c.00
3-6
Storage Configuration
wmtest can be used to test the capability of being able to fence the
server BUT it takes command line arguments. The real fencing
information is stored on the servers.
If wmtest works but fencing doesnt, check the fencing configuration
stored on each the nodes and in the membership partitions
/etc/opt/hpcfs/myFENCEidentity
3-7
hf836s c.00
3-7
Storage Configuration
Troubleshooting
What happens when the node cannot be fenced?
Successful fencing is critical to data integrity, to prevent uncoordinated
will wait forever, there is no timeout, and eventually it will appear like i/o
in the cluster has hung, until verification is received that the node has
been fenced, such as rebooting the node
Failures such as SAN ADM unable to get to switch interface, or wrong IP
address has been specified for Server Management port
If you are certain the node is down, enter the following command:
mx server markdown <server>
Use with caution as data corruption could result if the node still has access to
the storage
Informs Matrix Server the node has been fenced and to ignore it
3-8
hf836s c.00
3-8
Storage Configuration
Node 1
Node 2
Node 3
Votes: 2 3
Votes: 2
Quorum Disk
3-9
hf836s c.00
Node 4
3-9
Storage Configuration
Node 1
Node 3
Membership Partition(s)
(Server Registry)
Access to majority
of membership
partitions
Node 1
lock
Node 2
Node 3
Lowest IP
Node 4
3 - 10
hf836s c.00
Node 2
3 - 10
Node 4
hf836s c.00
3 - 11
Storage Configuration
Storage Configuration
Membership partitions
PolyServe uses a set of membership partitions to control access to the
SAN and to store the device naming database, which includes the global
device names for SAN disks imported into the matrix.
PolyServe.
3 - 12
hf836s c.00
3 - 12
Storage Configuration
3 - 13
hf836s c.00
3 - 13
Storage Configuration
3 - 14
hf836s c.00
3 - 14
Storage Configuration
mounts its first shared file system, and remains in the registry
until it un-mounts its last file system, or leaves the matrix
3 - 15
hf836s c.00
3 - 15
Storage Configuration
mpdump -v output
Current Product MP Version: 2
Membership Partition Version: 2
Membership Partitions:
20:00:00:80:e5:11:ed:75::0/1 (ONLINE)
20:00:00:80:e5:11:ed:75::1/1 (ONLINE)
20:00:00:80:e5:11:ed:75::2/1 (ONLINE)
Membership Partition Device Database (Version 1):
UID:20:00:00:80:e5:11:ed:75::0
Label:psd1
(state=0x1/mask=00000000)
UID:20:00:00:80:e5:11:ed:75::1
Label:psd2
(state=0x1/mask=00000000)
UID:20:00:00:80:e5:11:ed:75::2
Label:psd3
(state=0x1/mask=00000000)
UID:20:00:00:80:e5:11:ed:75::7
Label:psd7
(state=0x1/mask=00000000)
UID:20:00:00:80:e5:11:ed:75::8
Label:psd8
(state=0x1/mask=00000000)
Membership Partition Volume Database (Version 2):
VOL:psv2
(stripesize=4096K)
Set 0: SUBDEV: 20:00:00:80:e5:11:ed:75::7/0
size=10240000K
SUBDEV: 20:00:00:80:e5:11:ed:75::8/0
size=10240000K
Membership Partition Host Registry (Version 3):
Host ID: 10.12.10.110 fencestatus=0 fencetype=0
Fence ID:21:00:00:e0:8b:1c:d2:50::qlogicswitch1 state=0
Host ID: 10.12.10.109 fencestatus=0 fencetype=0
Fence ID:21:00:00:e0:8b:1c:90:4f::qlogicswitch1 state=0
Host ID: 10.12.10.103 fencestatus=0 fencetype=0
Fence ID:21:00:00:e0:8b:1c:9b:4f::qlogicswitch1 state=0
Host ID: 10.12.10.104 fencestatus=0 fencetype=0
Fence ID:21:00:00:e0:8b:1c:10:41::qlogicswitch1 state=0
3 - 16
hf836s c.00
3 - 16
Storage Configuration
hf836s c.00
3 - 17
Storage Configuration
Extended availability
A node can access the MP datastore even if it loses access to all the MPs, as long as
at least one node in the matrix has access to the MPs (the request is automatically
redirected to a node that can access the MPs).
If no node can access the MPs, a local read-only early-access datastore can be
read instead. A nodes early-access datastore is a consistent copy of the MP
datastore, but may not contain all updates.
3 - 18
hf836s c.00
3 - 18
Storage Configuration
SCL
uses the cluster-ID to verify that MPs belong to the cluster
RBAC
roles are stored in /rbac
mxlog
reads the cluster-ID and cluster description for inclusion in log messages
SNMP traps
enabled flag, event filter, and trap targets are stored in /notifiers
hf836s c.00
3 - 19
Storage Configuration
3 - 20
hf836s c.00
3 - 20
Storage Configuration
3 - 21
hf836s c.00
3 - 21
Storage Configuration
3 - 22
hf836s c.00
3 - 22
3 - 23
hf836s c.00
3 - 23
Storage Configuration
Storage Configuration
which loses (or has corrupted) that partition must be taken out
of service to effect repair/replacement
No ADM election can take place
3 - 24
hf836s c.00
3 - 24
3 - 25
hf836s c.00
3 - 25
Storage Configuration
Storage Configuration
hf836s c.00
3 - 26
hf836s c.00
3 - 27
Storage Configuration
Storage Configuration
3 - 28
hf836s c.00
3 - 28
Storage Configuration
Node 1
Node 2
/dev/psd1
Node 3
/dev/psd1
/dev/sda
Consistent device
naming across the
cluster
/dev/sda
/dev/psd1
/dev/sdd
LUN 1
Prevents access to
/dev/sdx devices
3 - 29
hf836s c.00
3 - 29
hf836s c.00
3 - 30
Storage Configuration
Storage Configuration
the matrix
3 - 31
hf836s c.00
3 - 31
Deport buttons
on the UI are
used to import
and deport
disks, not
volumes
3 - 32
hf836s c.00
3 - 32
Storage Configuration
3 - 33
hf836s c.00
3 - 33
Storage Configuration
Removes a disk
systems
A membership
partition
3 - 34
hf836s c.00
3 - 34
Storage Configuration
Storage Configuration
Management Console:
The Import Disks and Deport Disks windows display the UID,
vendor, model, and capacity for imported or unimported disks
These windows also display the FC switch used to access each
disk
sandiskinfo (VERY useful tool)
# sandiskinfo -ial
# sandiskinfo -ual
3 - 35
hf836s c.00
3 - 35
3 - 36
hf836s c.00
3 - 36
Storage Configuration
Storage Configuration
(Membership Disk)
6-6005-08B4-0008-00C9-0004-8000-0131-0000
Vendor:
HP HSV300
Capacity: 1024.00M
(PSMP/Active )
(Membership Disk)
6-6005-08B4-0008-00C9-0004-8000-0135-0000
Vendor:
HP HSV300
Capacity: 1024.00M
(PSMP/Active )
(Membership Disk)
6-6005-08B4-0008-00C9-0004-8000-0139-0000
Vendor:
HP HSV300
Capacity: 1024.00M
(PSMP/Active )
Disk: /dev/psd/psd4
Uid:
6-6005-08B4-0008-00C9-0004-8000-0149-0000
Vendor:
HP HSV300
Capacity: 51200.00M
(SUBDEV/psv1)
Disk: /dev/psd/psd5
Uid:
6-6005-08B4-0008-00C9-0004-8000-014D-0000
Vendor:
HP HSV300
Capacity: 51200.00M
3 - 37
hf836s c.00
(SUBDEV/psv2)
2009 Hewlett-Packard Development Company, L.P.
3 - 37
hf836s c.00
3 - 38
Storage Configuration
Storage Configuration
CVM overview
Matrix Server V3 introduced a Cluster Volume Manager that
hf836s c.00
3 - 39
Storage Configuration
Basic volumes
Basic volumes (psd)
Consist of a single disk or LUN that has been imported into the
matrix
A PSFS file system is then created directly on the disk, partition,
or LUN
If the underlying lun can be expanded in the array then the file
system can be extended to use the additional space,
This may be an offline operation
3 - 40
Basic volumes
Volumes are used to store PSFS file systems.
hf836s c.00
3 - 40
Storage Configuration
Dynamic volumes
Dynamic volumes (psv)
Created by the Matrix Server Volume Manager
Can include one or more subdevices, such as disks, disk
partitions, or LUNs that have been imported into the matrix
A single PSFS file system can be placed on each dynamic
volume
Can add more subdevices to a dynamic volume as necessary
and extend the file system on the volume online
3 - 41
hf836s c.00
3 - 41
Storage Configuration
Concatenated volume
.
Each sub device is completely filled
before using next one
3 - 42
hf836s c.00
300 GB
(Full)
100 GB
(Full)
200 GB
3 - 42
Storage Configuration
100 GB
100 GB
100 GB
100 GB
300
100 GB
GB
100 GB
Dont Do This!
Data Striped
across multiple
LUNs.
Results
in Inconsistant
Performance
3 - 43
hf836s c.00
3 - 43
Grow striped
volumes by
adding uniform
stripe sets for
consistent
performance
Storage Configuration
Striped volume
150 GB
100 GB
200 GB
hf836s c.00
3 - 44
Wasted
Space
Storage Configuration
3 - 45
hf836s c.00
3 - 45
Storage Configuration
size).
3 - 46
hf836s c.00
3 - 46
size
Click OK or Apply
Apply keeps window open
psv name is assigned to the
dynamic volume
starting point
3 - 47
hf836s c.00
3 - 47
Storage Configuration
Storage Configuration
Unstriped Volume is
concatenated
Suboptimal Volume has
been extended and includes
more than one stripe set, the
volume may or may not have
been extended with a stripe
set
3 - 48
The volume properties can then be displayed from the Management Console
Stripe state will be one of the following:
Optimal the volume has only one stripeset that includes all subdevices
Unstriped the volume is concatenated
Suboptimal the volume has been extended and includes more than one stripeset, the
first stripeset will be filled before write to the next stripeset begin.
hf836s c.00
3 - 48
Storage Configuration
The Extend Volume option allows you to add subdevices to an existing dynamic volume.
When you extend the volume on which a file system is mounted, you can optionally
increase the size of the file system to fill the size of the volume.
Note: The subdevices used for a striped dynamic volume are called a stripeset. When a
striped dynamic volume is extended, the new subdevices form another stripeset. If you want
the entire dynamic volume to be in the same stripeset, you will need to recreate the volume.
hf836s c.00
3 - 49
Storage Configuration
Implement striping on a
concatenated volume, or
Place all subdevices in the
same stripe set if a striped
dynamic volume has been
extended
Volume Manager first
3 - 50
hf836s c.00
3 - 50
Storage Configuration
created directly on an
imported disk partition or
LUN (a basic volume),
convert the basic volume to
dynamic volume
3 - 51
hf836s c.00
3 - 51
Storage Configuration
3 - 52
When a dynamic volume is destroyed, the file system on that volume is also destroyed.
Before destroying a dynamic volume, be sure that the file system is no longer needed or
has been copied or backed up to another location.
The file system must be unmounted when you perform this operation.
On the Destroy Dynamic Volume Windows, select the volume to destroy then click Ok or
Apply
hf836s c.00
3 - 52
Storage Configuration
3 - 53
hf836s c.00
3 - 53
Storage Configuration
underlying disks
Will not destroy the file
system or underlying data
3 - 54
hf836s c.00
3 - 54
3 - 55
hf836s c.00
3 - 55
Storage Configuration
Storage Configuration
Unimportable Volumes
Unimported volumes may be IMPORTABLE or
TRUNCATED
One or more subdevices are smaller than the size recorded
in the subdevice signature
MISSING
One or more subdevices are missing
3 - 56
hf836s c.00
3 - 56
Storage Configuration
Subdevice re-use
The GUI will prompt before reusing a subdevice that
3 - 57
hf836s c.00
3 - 57
Storage Configuration
3 - 58
hf836s c.00
3 - 58
59
hf836s c.00
2009 Hewlett-Packard
Version 7.4 Development
HP Restricted
Company, L.P.
3 - 59
Storage Configuration
hf836s c.00
3 - 60
Storage Configuration
Module 4
hf836s c.00
HP Restricted
2009 Hewlett-Packard Development Company, L.P.
The information contained herein is subject to change without notice
hf836s c.00
4-1
Objectives
After completing this module you should be able to:
Configure file systems
Mount file systems
Perform file system checks
Configure user and group quotas
Create snapshots
Describe and Configure Replication
Explain Multi-Path IO software
4-2
hf836s c.00
4-2
hf836s c.00
4-3
4-4
Select Storage Filesystem New or click the Add Filesystem icon on the toolbar
The Label field identifies the file system
The storage extents (partitions) that are not currently in use are displayed in the
Available Extents area
Note: the Create a Filesystem window identifies disk partitions by their MxS names,
such as psd1p2. To match these names to their local Linux names, open the Get
Local Disk Info window
From the command line
mx fs create [--size <KB>] <filesystem> <storageExtent>
mkpsfs <device> [<size-in-blocks>]
mkfs t psfs ...
hf836s c.00
4-4
4-5
hf836s c.00
4-5
group quotas
4-6
hf836s c.00
4-6
hf836s c.00
4-7
Ordered/Unordered
Defines the order in which
metadata and data blocks are
written to the f/s
Cluster-coherent locks
Use with caution, affects only
fcntl() locks. If this is disabled
then record locking must be
implemented a different way
to ensure data integrity
4-8
hf836s c.00
4-8
In the past with the PolyServe file system, every file used at
least one metadata block and 1 file system block
on a file system built with 8k blocks a 1kB file would consume 16kB of
the file system, an 8k metadata block and an 8k data block.
4-9
hf836s c.00
4-9
Features: SMALL_FILES
4 - 10
1.
2.
The old psfs.ko should fail to mount a new file system with this feature bit on. The
kernel log will have a message like this: "The on-disk revision level of filesystem psd4
is newer than that supported by this version of PSFS. You may need to upgrade to a
newer version of PSFS in order to access this filesystem.
Once enabled on an FS, no way to revert: The intention is to always have feature on. The
new 3.7.0 psfs.ko driver must always undertand file data stored using the new
model
hf836s c.00
4 - 10
|e/cn|
key
| |need|
--------------------------------------------------------------------------------------------------------------------------------------------------| 0|0 8358 0x0 SD(0), len 48, loc 4048 free space 65535, fsck need 0, format new|
(SD), mode -rwxrwxrwx, size 12, nlink 1, mtime 02/26/2009 10:28:55 blocks 8, uid 0, gid 0
---------------------------------------------------------------------------------------------------------------------------------------------------| 1|0 8358 0x1 IND(1), len 4, loc 4044 free space 0, fsck need 0, format new|
1 pointer
[ 8359]
===================================================================
4 - 11
hf836s c.00
4 - 11
|e/cn|
key
| |need|
---------------------------------------------------------------------------------------------------------------------------------------------------| 0|0 8358 0x0 SD(0), len 48, loc 4048 free space 65535, fsck need 0, format new|
(SD), mode -rwxrwxrwx, size 12, nlink 1, mtime 02/27/2009 10:00:20 blocks 8, uid 0, gid 0
----------------------------------------------------------------------------------------------------------------------------------------------------| 1|0 8358 0x1 DRCT(2), len 3976, loc 72 free space 65535, fsck need 0, format new|
"hello world\n"
4 - 12
hf836s c.00
4 - 12
4 - 13
hf836s c.00
4 - 13
4 - 14
These are highlights from results that Fitsum and Rob have sent to me, contact them for details
and official latest Guinness testing numbers.
The best 38X better results were obtained running iozones fileop with 4 million files, 4K
blocksize, quota enabled, Rhek5.2, single node (RAM 16K)
hf836s c.00
4 - 14
500
400
Series1
300
Series2
200
100
0
1
11
13
15
17
19
21
23
25
27
29
31
33
Thousands of files
4 - 15
hf836s c.00
4 - 15
1400
1200
1000
Series1
800
Series2
600
400
200
0
1
11
13
15
17
19
21
23
25
27
29
31
33
Thousands of files
4 - 16
hf836s c.00
4 - 16
Psfscheck.exe
functionality.
Psfsformat.exe
Psfsinfo.exe
Psfsdebug.exe
4 - 17
hf836s c.00
4 - 17
4 - 18
hf836s c.00
4 - 18
is optimized for file systems that will house database data files by
supporting direct I/O
Disables file system buffering for I/Os (buffer cache bypass)
hf836s c.00
4 - 19
4 - 20
hf836s c.00
4 - 20
Persistent mounts
Used to ensure that file systems
4 - 21
hf836s c.00
4 - 21
4 - 22
hf836s c.00
(1 of )
4 - 22
(2 of 2)
<device>
The SMALL_FILES features was
explained earlier
4 - 23
hf836s c.00
4 - 23
4 - 24
hf836s c.00
4 - 24
(1 of 2)
(atime) automatically.
applications
4 - 25
hf836s c.00
4 - 25
(2 of 2)
/etc/var/polyserve/mxinit.conf file:
psfs_atime_enabled = 1
mxinit.conf file:
4 - 26
hf836s c.00
4 - 26
(1 of 2)
4 - 27
hf836s c.00
4 - 27
(2 of 2)
4 - 28
hf836s c.00
4 - 28
being evicted
being evicted
4 - 29
hf836s c.00
4 - 29
hf836s c.00
4 - 30
Quota overview
(1 of 3)
introduced
The PSFS file system supports both hard and soft file system
Soft limits provide a way to warn users (or groups) when they
4 - 31
Hard quotas
When a file owner reaches the hard limit, the file system will not allow the owner to
create files or to increase the size of an existing file
Any attempts to allocate more space will fail.
The file owner will need to remove files or reduce their size until the disk usage falls
below the hard limit.
Soft quotas
A soft limit is typically set below the hard limit and triggers the warning.
If you want to use soft limits, you will need to configure a warning mechanism such as
the Linux warnquota utility.
hf836s c.00
4 - 31
Quota overview
(2 of 3)
Hard quotas
When a file owner reaches the hard limit, the file system will not
allow the owner to create files or to increase the size of an existing
file
Any attempts to allocate more space will fail.
The file owner will need to remove files or reduce their size until the
disk usage falls below the hard limit.
Soft quotas
A soft limit is typically set below the hard limit and triggers the
warning.
If you want to use soft limits, you will need to configure a warning
mechanism such as the Linux warnquota utility.
4 - 32
Hard quotas
When a file owner reaches the hard limit, the file system will not allow the owner to
create files or to increase the size of an existing file
Any attempts to allocate more space will fail.
The file owner will need to remove files or reduce their size until the disk usage falls
below the hard limit.
Soft quotas
A soft limit is typically set below the hard limit and triggers the warning.
If you want to use soft limits, you will need to configure a warning mechanism such as
the Linux warnquota utility.
hf836s c.00
4 - 32
Quota overview
(3 of 3)
When you create a PSFS file system, you will need to specify
system.
hf836s c.00
4 - 33
Quota configuration
Check or uncheck Enable quotas
as appropriate.
4 - 34
hf836s c.00
4 - 34
Quota editor
4 - 35
The Management Console includes a quota editor that you can use to view quota
information and to set or change the hard and soft limits for specific users or groups on a
file system.
You can start the editor from the Quotas tab on the File System Properties window
Right-click on a File system, select Properties, and then click on Manage Quotas
You can also use the menus: Storage Manage Quotas at the top of Management
Console.
hf836s c.00
4 - 35
Quota searches
4 - 36
You can use the search feature on the left side of the quota editor to locate quota
information for specific users or groups.
If you are searching by name, the quota information must be in a database (such as a
password file or LDAP database) that can be accessed from the server where the file
system is mounted.
The search locates the name in the database and matches it with the ID, which is the
value stored on the file system
hf836s c.00
4 - 36
Adding quotas
4 - 37
(1 of 2)
hf836s c.00
4 - 37
Adding quotas
4 - 38
(2 of 2)
When the Add Quota dialog appears, select the appropriate file system and set the
quota limits.
Any existing quota limits on the file system will be overwritten.
hf836s c.00
4 - 38
system you can remove the quotas for that user or group.
Select the user (or group) on the Quotas dialog and then click
the Delete icon on the toolbar.
Note: The quotas cannot be removed if the user or group has
4 - 39
hf836s c.00
4 - 39
4 - 40
hf836s c.00
4 - 40
4 - 41
hf836s c.00
4 - 41
Note: You do not need to install this RPM to use quotas with
4 - 42
hf836s c.00
4 - 42
4 - 43
hf836s c.00
4 - 43
hf836s c.00
4 - 44
Array-Based Snapshots
4 - 45
hf836s c.00
4 - 45
storage arrays, the latest version of the HP StorageWorks Scripting System Utility
(SSSU) must be installed on all servers in the cluster. Also, the latest version of
CommandView EVA software must be installed on your Management Appliance.
Be sure that your versions of SSSU and CommandView EVA are consistent. The
SSSU utility must be renamed, or linked, to /usr/sbin/sssu, and must be
executable by all users. To locate this software, contact your HP representative.
SANtricity Storage Manager client software must be installed on all servers in the
cluster. Also, the latest version of firmware must be installed on your storage array
controllers. To locate this software and firmware, contact your Engenio
representative.
4 - 46
hf836s c.00
4 - 46
(MSA2000) storage arrays, the latest version of firmware must be installed on the
array controllers and the SSH Command Line Interface (CLI) service must be
enabled on the array controllers. Also, note that a MSA2000 snapshot license is
required. Only the file systems located on Master Volumes (not Standard Volumes)
are snapshot capable.
HP XP storage arrays
To take hardware snapshots on HP StorageWorks XP24000, XP20000,
4 - 47
hf836s c.00
4 - 47
4 - 48
hf836s c.00
4 - 48
4 - 49
hf836s c.00
4 - 49
Creating a snapshot
(1 of 2)
To create a snapshot:
Select the file system on the
Management Console
Right-click and select
Create Snapshot.
The file system must be
mounted.
4 - 50
hf836s c.00
4 - 50
Creating a snapshot
(2 of 2)
4 - 51
hf836s c.00
4 - 51
Mounted snapshots
Mounted snapshots appear on the Management Console
beneath the entry for the file system because they still need
the original file system, they are really just deltas from the
original
name.
4 - 52
hf836s c.00
4 - 52
Mounted snapclones
A snapclone appears as its own filesystem mount in the
4 - 53
hf836s c.00
4 - 53
4 - 54
hf836s c.00
4 - 54
Delete a snapshot
Storage arrays typically limit the number of snapshots that
an existing snapshot.
Also, if you want to destroy a file system, you will first need
To delete a snapshot:
Select the snapshot on the Management Console
Right-click and select Delete.
To delete a snapshot from the command line, type the
following:
hf836s c.00
4 - 55
hf836s c.00
4 - 56
4 - 57
hf836s c.00
4 - 57
4 - 58
hf836s c.00
4 - 58
IS NOT
Database backup
Aimed at file-oriented
replication
Designed to get replication
into a customers site with low
pain
Realtime
4 - 59
hf836s c.00
4 - 59
writes summary info about changes to logs on the admin file system
Sentinel node use rsync to detect changes to the file system and
4 - 60
hf836s c.00
4 - 60
Sentinel
takes information about changes from log file
reads data and compares with target
sends to replication partner.
hf836s c.00
4 - 61
01
010110
adminfs
PSFS
4 - 62
hf836s c.00
PSFS
4 - 62
2.
3.
4.
4 - 63
hf836s c.00
4 - 63
4 - 64
hf836s c.00
4 - 64
Set up the ssh configuration for replication on both the source and
destination clusters, it uses ssh t communicate with the destination.
You can create a custom key pair and then install that key pair on the
source and destination clusters. Use one of these methods:
Run rplkeys -c
creates the custom key pair, adds it to the mxds datastore, and publishes it for
replication use.
The rplkeys -c and -i commands attempt to publish the key on the destination
4 - 65
hf836s c.00
4 - 65
hf836s c.00
4 - 66
4 - 67
hf836s c.00
4 - 67
4 - 68
hf836s c.00
4 - 68
Best practices
If possible, structure directories to reduce # of watch points
Make sure you need to replicate the data thats flowing, dont just default
to replicating everything
Dedicate a node to being the sentinel under normal operation
Network bandwidth matters so use a dedicated network for replication
Troubleshooting
Watch CPU load, memory utilization on sentinel node; if it gets too high,
you may get problems.
69
hf836s c.00
4 - 69
Replication Commands
rplmonitor
Start and stop replication system.
Typically run by mxinit, but useful for starting/stopping replication
without node reboot.
rplstatus
Display replication state (configured, running, stopped, etc.) and
current sentinel node (also scripting extensions)
rplcontrol
Control replication state (start, stop, pausetransport, etc.)
rplkeys
Setup up replication SSH keys
70
hf836s c.00
4 - 70
Replication Troubleshooting
rpl_create_hr
Convert rplwatch binary log files to text
Typically run by mxinit, but useful for starting/stopping replication
without node reboot.
rplctldump
Convert history file to readable text
Useful for seeing exactly whats been replicated.
rpl_versions.sh
Verify OS and HP Scalable NAS versions on source and
destination are the same
71
hf836s c.00
4 - 71
Replication Logs
The replication log files created by rplwatch are located in the /_adminfs file system under
the directory /replication/logs. The following types of logs are created:
Path logs. These log files contain the inodes and absolute paths of the directories included
in the replica set. Replication uses this information internally to find the absolute paths for
modified files in the directories.
Change logs. These log files contain a history of the watched files and changes to those
files during a single replication interval. There is a change log for each node. At the end of
a replication interval, these logs are merged into one change log that specifies all of the files
that are to be transferred to the destination cluster. (If a file already exists on the destination
cluster, only the changes in the file will be transferred.)
Delete logs. These log files contain a history of the files that were deleted during a single
interval. There is a delete log for each node. At the end of a replication interval, these logs
are merged into one delete log that specifies all of the files that are to be deleted from the
destination cluster.
The log files are in a binary format and cannot be viewed with a text editor. Use the
rpl_create_hr command as follows to convert the files into a readable format.
4 - 72
hf836s c.00
4 - 72
4 - 73
hf836s c.00
4 - 73
steps:
# rplcontrol -s stop
Use a file transfer method (for example, scp, ftp, tar, cpio) to copy the files back
# rplcontrol -s restart
Note: Replication does not provide tools to analyze or report the
other direction. There is not an advantage over scp if there is total data loss on
the source. There is an advantage if the data loss is large but not total, as some
files will not need to be replicated.
4 - 74
hf836s c.00
4 - 74
The command can be run as part of a monitor script using different arguments
#rplstatus status
This will return zero (success) if replication is running on the node, this should succeed
on every node
#rplstatus sentinel
This should return zero (success) if replication is the sentinel is up and running, should
on succeed on the sentinel
The next slide shows these command running a s a custom device monitor being used to
monitor replication in the cluster
4 - 75
hf836s c.00
4 - 75
4 - 76
hf836s c.00
4 - 76
4 - 77
hf836s c.00
4 - 77
hf836s c.00
4 - 78
Multipath I/O
Used to eliminate single points of failure
Requires LUN to be presented with the same WWNN:LUN# at
the same time on all storage controllers
It is expected that the requires third party MPIO software will be
installed and configure
A matrix can have
Multiple FC switches, multiple FC ports per server and multiported SAN disks
Requires third party array-specific software to support MPIO
regardless of the active/passive or active/active nature of the
array
4 - 79
hf836s c.00
4 - 79
Multipath I/O
HP PolyServe for Linux has a built in mpio driver, mxmpio, but is not
4 - 80
hf836s c.00
4 - 80
hf836s c.00
4 - 81
Cluster.
Partition disks before enabling DM_MPIO or a reboot
may be required to recognize new partitions.
Refer to HP Scalable NAS Documentation
Refer to OS Distribution Documentation.
http://www.redhat.com/docs/manuals/enterprise/RHEL-5manual/en-US/RHEL510/DM_Multipath/index.html
http://www.novell.com/documentation/sles10/stor_evms/index.ht
ml?page=/documentation/sles10/stor_evms/data/multipathing.htm
l
4 - 82
The cluster software expects that if the device mapper multipath is to be used, that it is installed
and configured before starting the cluster. Our install and upgrade documents talk about how to
do this. Also, there is documentation from the OS distribution that tells how to set up device
mapper multipath.
Our documentation only mentions using the HP device mapper enablement package to get the
required multipath.conf file and setting for HBA parameters. The correct multipath.conf settings for
other vendor's storage would have to come from the vendor.
MP disk partitions should be created before starting device mapper multipath. Device mapper
and partitions don't work particularly well with each other. If a disk has partitions already on it,
that is fine. If you want to put partitions on a disk that is already controlled by dm-mpio, then it is
more trouble. The distribution documentation says to reboot after creating partitions. This is
because re-reading the partition data when the device is controlled by the device mapper doesn't
work. This is why, to make things easier, partitions should be added to MP disks before
enabling device mapper multipath.
hf836s c.00
4 - 82
4 - 83
hf836s c.00
4 - 83
Cluster.
Partition disks before enabling DM_MPIO or a reboot
may be required to recognize new partitions.
Refer to HP Scalable NAS Documentation
Refer to OS Distribution Documentation.
http://www.redhat.com/docs/manuals/enterprise/RHEL-5manual/en-US/RHEL510/DM_Multipath/index.html
http://www.novell.com/documentation/sles10/stor_evms/index.ht
ml?page=/documentation/sles10/stor_evms/data/multipathing.htm
l
4 - 84
The cluster software expects that if the device mapper multipath is to be used, that it is installed
and configured before starting the cluster. Our install and upgrade documents talk about how to
do this. Also, there is documentation from the OS distribution that tells how to set up device
mapper multipath.
Our documentation only mentions using the HP device mapper enablement package to get the
required multipath.conf file and setting for HBA parameters. The correct multipath.conf settings for
other vendor's storage would have to come from the vendor.
MP disk partitions should be created before starting device mapper multipath. Device mapper
and partitions don't work particularly well with each other. If a disk has partitions already on it,
that is fine. If you want to put partitions on a disk that is already controlled by dm-mpio, then it is
more trouble. The distribution documentation says to reboot after creating partitions. This is
because re-reading the partition data when the device is controlled by the device mapper doesn't
work. This is why, to make things easier, partitions should be added to MP disks before
enabling device mapper multipath.
hf836s c.00
4 - 84
Download the HPDM Multipath Enablement Kit for HP StorageWorks Disk Arrays
v4.2.0 available
at http://www.hp.com/go/devicemapper.
/tmp/HPDMmultipath).
#cd /tmp/HPDMmultipath
#cd HPDMmultipath-4.2.0
Verify that the directory contains README.txt, COPYING, INSTALL, bin, conf,
SRPMS, and
docs directories.
4 - 85
hf836s c.00
#./INSTALL
4 - 85
4 - 86
hf836s c.00
4 - 86
*
*
********************************************************************************
Installation completed successfully!
[root@poly2 HPDMmultipath-4.2.0]#
4 - 87
hf836s c.00
4 - 87
Path grouping policies: Paths are coalesced based on the following path-
grouping policies:
Multibus
All paths are grouped under a single path group
Group by serial
Paths are grouped together based on controller serial number
Failover only
Provides failover without load balancing by grouping the paths into individual path groups
I/O load balancing policies: Provides weighted Round Robin load balancing policy within a path group
I/O load balancing policies: Provides weighted Round Robin load balancing
hf836s c.00
4 - 88
"HP|COMPAQ"
product
"HSV1[01]1 \(C\)COMPAQ|HSV[2][01]0|HSV300"
path_grouping_policy
group_by_prio
getuid_callout
"/sbin/scsi_id -g -u -s /block/%n"
path_checker
tur
path_selector
"round-robin 0"
prio_callout
"/sbin/mpath_prio_alua /dev/%n"
rr_weight
uniform
failback
immediate
hardware_handler
no_path_retry
rr_min_io
"0"
12
100
4 - 89
hf836s c.00
4 - 89
multipath.conf entry - XP
For XP
device {
vendor
"HP"
product
"OPEN-.*"
path_grouping_policy
multibus
getuid_callout
"/sbin/scsi_id -g -u -s /block/%n"
path_selector
"round-robin 0"
rr_weight
uniform
path_checker
tur
hardware_handler
"0"
failback
immediate
no_path_retry
12
rr_min_io
1000
4 - 90
hf836s c.00
4 - 90
hf836s c.00
4 - 91
4 - 92
hf836s c.00
4 - 92
# resizepsfs /dev/psd/psd6p4
(to use all available space)
hf836s c.00
4 - 93
braces:
4 - 94
hf836s c.00
4 - 94
(1 of 2)
4 - 95
hf836s c.00
4 - 95
(2 of 2)
4 - 96
96
hf836s c.00
4 - 96
Before that server can reboot, the port will need to be manually
unfenced.
The preferred fencing method when booting from SAN is to use
Flexible Server Fencing.
It is possible to use a switch that is not managed by Matrix
hf836s c.00
4 - 97
(1 of 2)
4 - 98
hf836s c.00
4 - 98
(2 of 2)
4 - 99
hf836s c.00
4 - 99
When this lower-level OLI support is in place, inserting a new disk will
4 - 100
hf836s c.00
4 - 100
The replacement switch must be the same model as the original switch
and must have the same number of ports.
The FC connectors must be reinserted in the same location on the new
switch.
PolyServe MxS must be stopped on any servers that are connected
only to the switch to be replaced.
If these conditions are not met, you will not be able to perform online
4 - 101
hf836s c.00
4 - 101
hf836s c.00
4 - 102
4 - 103
103
hf836s c.00
2009 Hewlett-Packard
Version 7.4 Development
HP Restricted
Company, L.P.
4 - 103
hf836s c.00
4 - 104
Cluster administration
Module 5
hf836s c.00
hf836s c.00
5-1
Cluster administration
Objectives
At the end of this module, students will be able to:
Understand the security architecture of the HP PolyServe
Software
Authentication
Role Based Access Control
Understand the various management interface points of the
5-2
hf836s c.00
The CLI
SNMP
Notifiers
Performance Dashboard
2009 Hewlett-Packard Development Company, L.P.
5-2
hf836s c.00
Cluster administration
5-3
Cluster administration
5-4
hf836s c.00
5-4
Cluster administration
5-5
Servers Tab
This gives a server view of the matrix, including the network interfaces
Virtual Hosts Tab
Show all vhosts in matrix, can drill down into members and services
Applications Tab
Shows the application monitors configured in the matrix
Filesystems Tab
Shows all psfs file systems in the matrix
Notifiers Tab
Shows all notifiers configured in the matrix
hf836s c.00
5-5
Cluster administration
Applications tab
Application monitors
Filesystems tab
All PSFS file systems in the
matrix
5-6
hf836s c.00
5-6
Cluster administration
When you begin the drag operation the cursor will change to a
circle with a bar through it meaning that the current mouse location
does not allow drops.
5-7
hf836s c.00
5-7
Cluster administration
5-8
hf836s c.00
5-8
Cluster administration
cluster software
versions running
on each node,
especially during
rolling upgrades
5-9
hf836s c.00
5-9
dialog a
consolidated view
of the relationship
between LUNs,
Volumes, and File
Systems
5 - 10
hf836s c.00
5 - 10
Cluster administration
Cluster administration
.matrixrc file
The mx utility can be used both interactively and in scripts.
5 - 11
hf836s c.00
5 - 11
Cluster administration
mx syntax
mx [mx_options] class command [command_options]
--help Displays a command summary.
The mx_options affect an entire mx command session. The
options are as follows:
--matrix <matrix> Specifies the matrix that you want to connect with.
matrix can be any node in the matrix.
--config <file> Specifies the configuration file to be consulted for
server, user, and password information. The file must have the
same format as matrixrc
5 - 12
hf836s c.00
5 - 12
Cluster administration
in the mx command.
5 - 13
hf836s c.00
5 - 13
Cluster administration
5 - 14
hf836s c.00
mx server add|update|delete|enable|disable|status|dump|read_license
mx vhost add|update|move|delete|status|enable|disable|dump
mx service add|update|delete|status|enable|disable|clear|dump
mx device add|update|delete|enable|disable|status|clear|dump
mx netif enable|disable|admin|noadmin|status|add|update|delete
mx notifier add|update|delete|enable|disable|status|test|dump
mx matrix destroy|dump|status|log
mx sleep
mx disk import|deport|status|dump
mx fs
create|resize|destroy|recreate|showcreateopt|mount|unmount|status|dump
getdriveletters|assignpath|unassign|queryassignments
mx alert status
mx snapshot create|destroy|showcreateopt
mx dynvol
mx server markdown
mx file <filename>
2009 Hewlett-Packard Development Company, L.P.
5 - 14
hf836s c.00
5 - 15
Cluster administration
Cluster administration
Authentication
Feature overview
Web browser and mxconsole authenticate as OS users and no longer
authenticate to the built-in admin user mxpasswd concept of users.
Pswebsrv supports only https on port 6771
MxS 3.5 mxconsole will NOT be able to connect to MxS 3.7
MxS 3.7 mxconsole can be used to connect to MxS 3.5
Password Authentication
.matrixrc, mx/mxconsole --user and --password, via GUI as user dialog or
when prompted
Apache module mod_authz_pmxs utilizes mxauthpw which authenticates with
PAM or if PAM not installed authenticates against shadow passwords.
MxS does NOT configure PAM, it merely uses the existing OS PAM configuration.
Best practices
Standard security best practices, i.e. dont logon as root
Saved passwords are encrypted in the .matrixrc as usual, but make sure this file is
protected
5 - 16
hf836s c.00
5 - 16
Cluster administration
Authentication
Caveats, limitations, known bugs
Troubleshooting
apache_access.log
401 need to authenticate and/or handshaking
200 authenticated successfully
apache_error.log
[Tue Feb 24 14:41:47 2009] [error] [client 10.10.211.6] AuthExtern
pmxs_auth [/opt/polyserve/lib/apache/bin/mxauthpw]: Failed (10) for user
foo
[Tue Feb 24 14:41:47 2009] [error] [client 10.10.211.6] user foo:
authentication failure for "/cgi-bin/pmxs/pulselet": Password
Mismatch
For unknown users or typo in username the error will still be reported as
Password Mismatch
Unfortunately, when you enter a utf8 username of - it prints as ???????? in the apache_error.log so no L18N support in the error messages
5 - 17
hf836s c.00
5 - 17
hf836s c.00
5 - 18
Cluster administration
Cluster administration
5 - 19
hf836s c.00
5 - 19
Cluster administration
Authenticated user access token contains user ID and associated group IDs
Used to determine which roles are associated with the
user
Will determine which rights are allowed/denied.
mxconsole configuration changes are audited (see mcs tool or GUI View Events)
hf836s c.00
5 - 20
Cluster administration
Without SAN access, roles in MxDS may become stale through the
early-access/off-line cache
perform the RBAC authorization check and have their actions audit
logged
5 - 21
hf836s c.00
5 - 21
hf836s c.00
5 - 22
Cluster administration
Cluster administration
3.7
Administrator Account
admin
Read-Only Guest
Any non-Administrators
member in a deny-all role
Password Management
Custom Authorization
N/A
RBAC
Single Sign-on
N/A
N/A
.matrixrc
Optional
Optional
Encryption
Some CGIs
Password Challenge
MD5
none
5 - 23
hf836s c.00
5 - 23
Use OS utilities
Cluster administration
hf836s c.00
5 - 24
Add
5 - 25
hf836s c.00
5 - 25
Cluster administration
Role Properties
Name
5 - 26
hf836s c.00
5 - 26
Cluster administration
Role Members
Assign Accounts to a Role
hf836s c.00
5 - 27
Cluster administration
My Rights
View the rights of the currently
logged in user
5 - 28
hf836s c.00
5 - 28
Cluster administration
Cluster administration
Export/Import
The import and export features can used if you will be
5 - 29
hf836s c.00
5 - 29
hf836s c.00
5 - 30
Cluster administration
Cluster administration
Logging
v3.7 introduces a new message catalog
Messages have unique IDs
Design considerations
5 - 31
hf836s c.00
5 - 31
Cluster administration
5 - 32
hf836s c.00
5 - 32
5 - 33
hf836s c.00
5 - 33
Cluster administration
Cluster administration
Logging
Best practices
Troubleshooting
General
Dont view the matrix log from a node that has MxS stopped
Events are copied to /var/log/messages immediately
If you need to see events during (early) startup or (late) shutdown, look in
/var/log/messages.
Matrix.log file still exists on Linux
Some event details omitted
Events not written there until mxlogd starts
Matrix log can be dumped or viewed locally with mcs select command.
Use mcs select non-customer to include messages hidden from MxS UI
Use cgi to dump in XML
Notifiers
Test scripts and email filters on specific message IDs with mcs log t i<event-id> .
These go to notifiers, but not to the matrix log.
Sends a single event from a single node, unlike the test GUI button which sends one event
from every running node (to test all the paths)
Cluster admin can omit -t, and inject events indistinguishable from real ones
5 - 34
hf836s c.00
5 - 34
Cluster administration
5 - 35
hf836s c.00
5 - 35
Cluster administration
Logging
Caveats, limitations, know bugs
The message catalog contains all messages used in any feature on any
platform
Messages may appear in the filter dialog that cant actually happen in a specific
installation.
5 - 36
hf836s c.00
5 - 36
5 - 37
hf836s c.00
5 - 37
Cluster administration
Cluster administration
Alerts
Overview
Alerts represent a persistent noteworthy condition in the system
Alerts persist until:
The condition they represent ends
The object the alert refers to (or that objects container) is deleted (or disabled)
Alerts cannot be cleared manually
Consider an email notifier instead
Alerts all have unique IDs, defined in the message catalog
Events are logged at the beginning and end of every alert
The beginning event for alert n has ID n, and the end event has ID n+1
There is one authority in the 3.7.0 cluster for the current state of all alerts
A UI gets the current state of all alerts when it connects
5 - 38
hf836s c.00
5 - 38
Alert Viewer
5 - 39
hf836s c.00
5 - 39
Cluster administration
Cluster administration
Alerts
Caveats, limitations, know bugs
3.5.x to 3.7.0 rolling upgrade issues
The alert authority doesnt exist until all nodes are running 3.7.0
3.7.0 nodes will only know about their own alerts (and the sanpulse alerts, which
take a different path) until the rolling upgrade is complete
Since alert events are logged by the alert authority, this doesnt happen until the
rolling upgrade is complete and a 3.7.0 alert authority exists
3.7.0 alerts will not be displayed on UIs attached to 3.5.x nodes
so the underlying problems (that dont cause failovers) may go undetected
during a prolonged rolling upgrade
but when the customer does it anyway, you can see the
active alerts on a 3.5.0 node by connecting a UI directly
to it
Sanpulse alerts from 3.7.0 will be displayed on 3.5.x consoles, but wont be readable
there
3.5.x doesnt understand the 3.7.0 alert encoding, and displays them as the raw
base 64 encoded alert structure (gibberish to users)
Any 3.7.0 node will display them correctly
5 - 40
hf836s c.00
Your MPs are too small for 3.7.0 is a very common 3.7.0 alert to see during a
rolling upgrade
2009 Hewlett-Packard Development Company, L.P.
5 - 40
hf836s c.00
5 - 41
Cluster administration
Cluster administration
Event Notifications
New event notification capabilities have been introduced
5 - 42
hf836s c.00
5 - 42
5 - 43
hf836s c.00
5 - 43
Cluster administration
the event notifier services (SNMP, email, script) and specify the matrix
events that will trigger them.
5 - 44
hf836s c.00
5 - 44
Cluster administration
5 - 45
hf836s c.00
5 - 45
Cluster administration
5 - 46
hf836s c.00
5 - 46
Cluster administration
5 - 47
hf836s c.00
5 - 47
Cluster administration
5 - 48
hf836s c.00
5 - 48
Cluster administration
5 - 49
hf836s c.00
5 - 49
Cluster administration
Cluster administration
#!/bin/bash
# Basic Notifier Script that sends email to root
# on the localhost. This will be the node that is
# the Group Leader at the time
exec mail -s
5 - 50
hf836s c.00
5 - 50
Cluster administration
Settings tab. This service runs a script when an event configured for
the service occurs. You can specify only one script. Data sent as xml
to STDIN and MxS-event-* environment variables
5 - 51
hf836s c.00
5 - 51
hf836s c.00
5 - 52
Cluster administration
SNMP Sub-Agent
Feature overview
Provides read-only access to cluster state/status information
Servers
serverTable, netifTable
Vhosts/SMs
vhostTable, vhstatTable,
DMs
devmonTable, dmstatTable,
File systems
filesystemTable, fsmountTable
Alerts
alertTable
svcmonTable, smstatTable
dmserverTable, dmvhostTable
5 - 53
hf836s c.00
5 - 53
Cluster administration
Cluster administration
SNMP Sub-Agent
Usage
To enable, add the following line to the etc/snmp/snmpd.conf
file:
dlmod mxsnmp /opt/hpcfs/lib/snmp/mxsnmp.so
[ OK ]
Starting snmpd:
[ OK ]
5 - 54
hf836s c.00
5 - 54
SNMP Sub-Agent
Example /etc/snmp/snmpd.conf file
5 - 55
hf836s c.00
5 - 55
Cluster administration
SNMP Sub-Agent
Example MIB browser output
5 - 56
hf836s c.00
5 - 56
Cluster administration
Cluster administration
SNMP Sub-Agent
Troubleshooting
First, try to isolate problem with log files:
/var/log/messages on target server
/var/opt/polyserve/mxsnmp.log on target server
If problem appears to be a MIB leaf object or MIB table
object problem:
hf836s c.00
5 - 57
hf836s c.00
5 - 58
Cluster administration
Cluster administration
Performance Dashboard
V3.7 creates a new and improved performance monitoring
capability
5 - 59
hf836s c.00
5 - 59
Cluster administration
Performance Dashboard
Create the administrative file system
The Performance Dashboard uses an administrative file system that
must be created on the shared storage. This file system is also used by
the replication feature and was described earlier in the presentation
5 - 60
hf836s c.00
5 - 60
Cluster administration
Performance Dashboard
Using the Performance Dashboard
The Performance Dashboard is a web-based application. To start the
dashboard, either click the Dashboard icon on the Management
Console toolbar or open a browser and enter the following URL,
where <node> is the server name or IP address of the node that will
run the dashboard.
https://<node>:6771/perfmon
The dashboard opens in the browser with a full view of the cluster to
which the connected node belongs (the Cluster Report). You will need
to authenticate to the dashboard by entering the fully qualified NTLM
(DOMAIN\User) or UPN (user@FQDN) credentials.
5 - 61
hf836s c.00
5 - 61
Performance Dashboard
5 - 62
hf836s c.00
5 - 62
Cluster administration
Performance Dashboard
5 - 63
hf836s c.00
5 - 63
Cluster administration
hf836s c.00
5 - 64
Cluster administration
hf836s c.00
5 - 65
Cluster administration
hf836s c.00
5 - 66
Cluster administration