5.3
Project
Project Planning
Planning
Network
Network Integration
Integration
PureFlex
PureFlex PowerVM
PowerVM Tuning
Tuning
SAN
SAN Integration
Integration
Initialization
Initialization
PureFlex
PureFlex upgrade
upgrade
FCoE
FCoE
FSM
FSM Discovery
Discovery and
and Inventory
Inventory
Automation
Automation Manager
Manager
SMIA
SMIA Discover
Discover Fabrics
Fabrics
Private Cloud
Using
Using VMControl
VMControl
42 customers
77 chassis
2011
5.3
Project Planning
IBM Power Systems Technical University Athens 2013
Project Planning
Ensure to do a project Kick-Off
Plan the right skill around the table
Don't assume that attendants already know; take the time to explain high level concepts
Define success criteria
Define time planning
When project is complex, better to take the time to produce a technical architecture document
To document expectations
To get stakeholders acceptance
Define responsibilities (RACI matrix)
2011
Network Planning
5.3
10
Mezzanine Card
I/O Module 1
I/O Module 2
11
Mezzanine Card
10Gb
pnic
4Gb
vnic
224 physical lanes
Up to 896 virtual lanes
12
Each I/O module in the Flex System Chassis is shown with two direct aggregations to a pair of
upstream Top-of-Rack (ToR) switches
This topology is suggested for environments in which hosts need network redundancy, but they are
not themselves performing any NIC teaming.
Offers complete network-level redundancy out of the chassis but if not balanced correctly, you lose
half of the available links and bandwidth due to Spanning Tree Protocol (STP) blocking them
13
Each I/O module in the Flex System Chassis connects directly to a single ToR switch through single or
aggregated links.
Topology is highly suggested when compute nodes use some form of NIC teaming.
This architecture with NIC teaming on the host, and Layer 2 Failover on the I/O modules, provides high
availability with no loops
No wasted bandwidth to spanning-tree blocked links.
14
15
Features
16
17
http://www.redbooks.ibm.com/technotes/tips0880.pdf
Copyright IBM Corporation 2013
18
VIOS1
S1_INT5A
S1_EXT1
en0 S1_INT5A
en1 S2_INT5A
S1_EXT2
S1_INT6A
S1_EXT3
S1_EXT4
VIOS2
en2 S1_INT6A
en3 S2_INT6A
P1-C36-L1-T1
-T2
SWITCH 1
S2_INT5A
S2_EXT1
S2_EXT2
S2_INT6A
S2_EXT3
S2_EXT4
SWITCH 2
19
VIOS1
S1_EXT1
S1_EXT2
S1_EXT3
S1_INT5A
en0 S1_INT5A
S1_INT5B
en1 S2_INT5A
S1_INT6A
en2 S1_INT6B
S1_INT6B
S1_EXT4
en3 S2_INT6B
VIOS2
en0 S1_INT5B
en1 S2_INT5B
en2 S1_INT6A
en3 S2_INT6A
P1-C34-L2-T1
-T2
P1-C36-L1-T1
-T2
SWITCH 1
S2_INT5A
S2_EXT1
S2_INT5B
S2_EXT2
S2_INT6A
S2_EXT3
S2_INT6B
S2_EXT4
SWITCH 2
20
VIOS1
S1_INT5A
S1_EXT1
P1-C34-L1-T1
-T2
en0 S1_INT5A
en1 S2_INT5A
S1_EXT2
S1_EXT3
S1_EXT4
DEV
SWITCH 1
PVID=1
S2_INT5A
S2_EXT1
SEA
S2_EXT2
S2_EXT3
S2_EXT4
en4
SWITCH 2
Switch 1
Switch 2
Backbone
Backbone
21
VIOS1
S1_INT5A
S1_EXT1
en0 S1_INT5A
en1 S2_INT5A
S1_EXT2
S1_EXT3
en2 S1_INT6B
S1_INT6B
S1_EXT4
en3 S2_INT6B
P1-C34-L1-T1
-T2
P1-C36-L2-T1
-T2
DEV
PVID=1
PVID=2
SEA
SEA
en4
en5
SWITCH 1
SWITCH 2
Backbone
Backbone
SWITCH 1
S2_INT5A
S2_EXT1
S2_EXT2
S2_EXT3
S2_EXT4
S2_INT6B
SWITCH 2
22
23
24
2011
SAN Planning
5.3
26
27
28
hdisk0
mpio
hdisk0
hdisk0
vscsi0
vscsi0
VIOS A
mpio/sddpcm
VIOS B
vhost0
fcs0
fcs1
SWITCH 3
vhost0
fcs0
fcs1
SWITCH 4
V7000
node1
V7000
node2
29
hdisk0
mpio/sddpcm
VIOS A
fcs0
fcs1
vfchost0
vfchost1
fcs0
fcs1
SWITCH 3
VIOS B
fcs2
fcs3
vfchost0
vfchost1
fcs0
fcs1
SWITCH 4
V7000
node1
V7000
node2
30
hdisk0
mpio/sddpcm
VIOS A
fcs0
fcs1
vfchost0
vfchost1
fcs0
fcs1
VIOS B
SWITCH /
PASST 3
fcs2
fcs3
vfchost0
vfchost1
fcs0
fcs1
SWITCH /
PASST 4
TOR
TOR
V7000
node1
V7000
node2
31
Zoning considerations
IBM Power Systems Technical University Athens 2013
vtSCSI
Zone VIOS with Storage
NPIV
Zone LPAR virtual WWPN with Storage
If you need LPM, ensure to zone both WWPN
lssyscfg -r prof -m Server-7895-42X-xxxxxx -F virtual_fc_adapters --filter
lpar_names=AIX
"""10/client/1/VIOS1/10/c050760586f60000,c050760586f60001/1"",
""11/client/2/VIOS2/11/c050760586f60002,c050760586f60003/1"",
""12/client/1/VIOS1/12/c050760586f60004,c050760586f60005/1"",
""13/client/2/VIOS2/13/c050760586f60006,c050760586f60007/1"""
Number of paths
Keep a reasonable number of paths (4 or 8 ok; 16 max)
32
2011
What to do to start
5.3
34
CMM IP address
192.168.70.100 if Flex
192.168.93.100 if PureFlex
35
36
Systems Management
IBM Power Systems Technical University Athens 2013
FSM
Network
SAN
Storage
Power nodes
PowerVM
(VIOS)
Intel nodes
IBMi
AIX
Chassis
(CMM)
Linux
KVM
VMWare VCenter
Win / Hyper-V
37
Hardware
Management
Network (GOLD)
eth0
eth1
FSM code
Active
CMM
I/O Module
1
KVM host
bond
Stand-by
CMM
bond
I/O Module
2
nodes
nodes
nodes
IMM /
FSP
I/O
nodes
nodes
Modules
INT-MGT
Customer Data
Network
(PURPLE / GREEN)
38
39
Switches
Chassis
Connects the Chassis Management Module (CMM), the IMM/FSP of the compute nodes,
switches in the I/O bays, the Flex V7000 storage and the Flex System Manager (FSM)
Uses a special Broadcom 5718-based management network adapter
FSM bonded pair of NIC ports connecting to ports 1 of both switch 1 and switch 2
Connects to the Internet for product updates and other related management functions
Note: The Eth0 or CMM management networks in multiple chassis are connected together
through the external ports of the CMMs in each chassis by using a GbE top-of-rack switch.
40
41
2011
5.3
After the Chassis has been managed it usually takes around 15 minutes for the inventory to be
collected,meaning the FSM's database will be updated with all the information regarding the
hardware elements present in the managed chassis. It's a good idea to wait for this process to
finish, if you have issues with patience (like I do) go to the Plugins tab and click the
active&scheduled jobs link, you should see the jobs running.
43
Usually access to the components should show OK, but just to make sure, have a look at the
Plugins tab, you should see immediately under the Inventory link if there are any systems with
No Access or Partial Access, in case there are make sure you request access to avoid any
future inventory collection or update issues.
44
If using V7000 storage you should discover&manage it from the FSM. If you have an internal Flex
System V7000 system you can use the Launch IBM Flex System V7000 EZSetup task and follow
the wizard, alternatively you can use the quick link, on the Additional Setup tab, Manage System
Storage. This will give you several possibilities to manage and monitor your storage subsystem, get
an overview of the disk usage, pools, hardware alerts, create and delete volumes.
45
46
First you might have to generate a public and private pair of rsa keys
ssh-keygen -t rsa -f $HOME/v7000_rsa, then using browse select the private key.
After the task is complete inventory will automatically be collected.
47
As inventory is a central component of great importance from the FSM's usability perspective I
recommend to do a complete collection after all the hardware has been discovered/managed,
my preference is the command line from a terminal/putty ssh session: smcli collectinv -p All
Inventory. Inventory collection is task that highly utilizes the database so I wouldn't
recommend doing too many things during this process.
48
2011
5.3
There are a number of methods to update the component firmware of your Flex System. Below
is the recommended order of operations for performing these depending on the hardware you
have.
For the latest information on firmware updating best practices, please see the Firmware Update
Guide at the link below.
https://www-947.ibm.com/support/entry/myportal/docdisplay?lndocid=MIGR-5091991
50
Obtaining Firmware
In addition to updating the firmware on the nodes, you should also update the CMM and
all the switch modules you have in the chassis. To make it easy, IBM has a single page
where you can easily access the latest firmware for all of the components.
http://www-01.ibm.com/software/brandcatalog/puresystems/centre/details?uid=S_PUREFLEX
51
Updating CMM
IBM Power Systems Technical University Athens 2013
To update the Chassis Management Module (CMM), log in to the CMM, click on Mgt Module
Management, and select Firmware.
If you have two CMMs, you only need to update the firmware on your primary CMM. Once
the update process is complete, and the CMM has rebooted, it will automatically update the
secondary CMM.
When you download the firmware package, you will need to extract the zip file. The update
process will require the file with the .uxp extension within that package.
52
53
To update the Qlogic FC3171 fiber channel switch, log in to (default user/pass is
USERID/PASSW0RD) and click on Switch > Load Firmware. Select the folder where the .bin
file firmware is located. You do not need to select the exact file. The switch will automatically
access the firmware file in that directory.
54
The networking switch has two components that need to be updated. The Boot code and the
Image code. Both the boot and image code are included in the single download package.
Typically we recommend updating the boot code first and then one of the image banks. It is
recommend that you keep one level back in one of the image banks.
Updating the firmware is as easy as clicking the browse button and then selecting the firmware
package.
55
If you have the Flex System Manager, updating chassis components is a easy process by
selecting the Update Chassis Components option and following the order provided. We
always recommend checking the Firmware Update Guide for the latest information and any
changes to the order listed in the Flex System Manager.
56
2011
5.3
To fully utilize the advanced features and capabilities of the FSM, like automatic workload
deployments, you should be able to read and write zoning information on your SAN
infrastructure.
Since most storage vendors developed their own language a need for commonality arose. The
Storage Networking Industry Association developed a standard, Storage Management
Initiative Specification, shortly SMI-S, to enable interoperable management of storage
systems.
If you have a Qlogic switch in full fabric mode all you need to do is discover the switch and
configure SNMP access. Qlogic switches have an SMI-S provider built in their firmware that
can be used to discover the zoning information. However due to current limitations this access
is read only, meaning you will not be able to deploy NPIV based workloads as the FSM can
not create new zones.
Brocade switches do not have this built in component thus an external tool, SMI-S provider is
required to be able to manage a Brocade based SAN infrastructure from the FSM. IBM
decided to preinstall this java based tool in the FSM, you have to to start and configure it as
shown in the next slide:
58
From the applications tab click Start under SMIA Configuration Tool:
59
Tip: Make sure you have a dns entry or you have the name in your local host file, credentials are
the same as for the FSM.
60
61
After changing to https Stop and Start form the Applications tab:
62
Do a Fabric discovery:
63
Click add, put the IP address of the Brocade switch and USERID/PASSW0RD by default:
64
After this you have to use the Discover Storage System task as with the V7000 but this time
select FC switch and provide the details:
65
66
After successfully configuring the IBM Fabric Manager, go to the Plugins tab, click on Resource
Explorer and click on the All Systems dynamic group. Type Farm in the search box, this is a
container object that represents all storage components known to the FSM, like the V7000
and SAN switches. Right click this object and select Collect Inventory. This should update the
FSM's database with all the zoning and SNMP information from the fabric switches, and also
with the information from the V7000. Since the fabric switches are now known to the FSM any
alerts generated by these will be received and displayed in the FSM.
67
2011
5.3
If you have Power nodes in your Chassis I would assume you will also have Virtual I/O Servers,
unless you want to use the Power nodes in full system partition mode, in which case more
specific tuning will have to be considered.
For quite some time now VIOS and AIX have the so called Common Agent installed by default.
Though it's not mandatory for basic management, I highly recommend making use of it's
features, especially for more advanced functions, like automating deployments of Virtual
Servers. To do this, after you have successfully installed and configured your VIOS, run a
discovery on the IP address of the VIOS, authenticate (request access), then collect inventory.
This will give the FSM access and visibility to the Operating System component of the Power
node, specifically of the VIOS. The procedure is the same with AIX systems.
69
Tips for creating the Virtual Server profile for the VIOS:
- use 1 or 2 dedicated processors in donating mode, this way you ensure the best locality and
availability of CPU and memory resources and also that the idle cycles can be used by the
other partitions. If you plan to have high network traffic without jumbo frames and vSCSI
storage you should have at least 2 CPU's, if you use jumbo frames and NPIV 1 CPU should
be enough. Memory should be between 6-8Gb to start and 12Gb maximum in case you have
to increase it later on the fly. Physical I/O can be split between 2 VIO servers depending on
the physical hardware capabilities, in a p460 for example you would typically have 4
mezzanine adapters, thus you can split 2-2 for creating a dual VIOS configuration. The
network adapters have 2 asics, thus even one adapter can be split in two, the 2 port fibre
channel adapters have one while the 4 port adapters have 2 asiscs. In the new p270 node
there is also a special adapter for splitting the on board SAS controller into 2 in which case
you could potentially install 2 separate VIO servers on local disks. This is not possible with
any other nodes as they only have 1 SAS controller.
70
Networking tips:
For high availability and redundancy reasons I suggest to consider creating etherchannel network
adapters on the VIOS. As LACP is natively supported this is what I prefer to use. It does imply
you will need VLAG and will have to create the LACP port groups on the chassis network
switches. I usually put INTAx on IOM1 with INTAx on IOM2 in one LACP group and INTBx
with INTBx in a different LACP group. If it's a full wide node with Dual VIOS I would use INTAy
with INTAy and INTBy with INTBy on the second VIOS. This configuration will give you
2x20Gbs adapters per VIOS. Then you have to create the etherchannel adapter making sure
you select type 802.3ad, and the en adapters corresponding to the ports you configured
previously. In most cases they would be en0 with en1 and en2 with en3 but it's best to double
check the location codes first.
Jumbo frames are very useful in reducing processing overhead and increasing throughput though
it should be used extensively throughout the infrastructure. Good news is the PureSystems
network switches all support this by default. Fine tuning for jumbo frames is quite an extensive
topic that usually has it's own session, please check the agenda for more information.
71
Next step is to create the SEA adapters. If you haven't done so while creating the virtual server
definition for the VIOS you should now create the required virtual network adapters,
depending on how many etherchannel adapters you had created in the previous step.I
suggest to have one virtual adapter for the management IP of the VIOS, then 1 or 2 for each
Shared Ethernet Adapter. If you are using Dual VIOS you will need at least 2 adapters/SEA,
one will be the trunk adapter, this one needs to have access to the external network, the
other one will be the control channel (heartbeat) adapter used for communication between the
2 VIO servers. The trunk adapter can be IEEE 802.1Q compatible and it can manage up to 20
VLANs. If you need to serve more VLAN's you will have to create additional adapters for each
bundle of 20.
This is an example command for creating the SEA:
mkvdev -sea ent9 -vadapter ent4 -default ent4 -defaultid 1 -attr
ha_mode=auto ctl_chan=ent5 largesend=1 large_receive=yes
Ent9 is the physical adapter, in our case the new etherchannel adapter, ent4 is the virtual trunk
adapter, the one that has the option to access external networks and ent5 is the control
channel adapter. Note that both ctl_chan virtual adapters, on VIOS1 and VIOS2 should have
the same unique PVID. If you plan to use jumbo frames, which I highly recommend also add
this parameter to the command: jumbo_frames=yes
72
Useful parameters:
For fibre channel adapters you should use dynamic tracking of devices and fast fail error
recovery, this will ensure that any missing paths will be noticed immediately and the multipath
I/O driver will initiate a failover and prevent loss of data. It is also recommended to install the
sdd-pcm driver that builds on the default mpio and adds extra functionality and usability.
chdev -dev fscsi0 -attr dyntrk=yes fc_err_recov=fast_fail -perm
Since the adapters are usually active these changes cannot be done on-line thus the -P
parameter which also implies you will have to reboot the VIOS to activate the changes.
For specific applications you might have to change additional parameters.
For vScsi disks in a dual VIOS setup you have to make sure to change the reservation policy
before assigning the disks to a partition:
chdev -dev hdisk# -attr reserve_policy=no_reserve
73
2011
5.3
VMControl is the first step into a Power systems based Cloud environment on the FlexSystem
platform, it provides the functionality to capture existing workloads and automatically deploy
new systems. Deployment includes the creation of the Virtual Server, creation of required
virtual network adapters, creation of required disks and their corresponding virtual adapters, all
the mappings plus the OS installation.
Given everything has been set-up correctly this process will complete under 5 minutes after a few
easy steps where you are required to provide the name, future IP address, hostname, dns
servers, etc.
The reason for being so quick is because it's based on flashcopy technology, either when using
V7000 for direct LUN allocation or VIOS Shared Storage Pools based virtual disks.
75
For this to work there a few key factors that have to be considered:
- the VIOS Common Agent has to be discovered and access has to be OK.
- the correct repository subagent has to be installed on the VIOS, and if using Shared Storage
Pools there is an additional subagent that you will have to install. This will ensure that you can
create an image repository to store all the metadata required to successfully catalog and
manage OS images that you will use in the future to deploy new workloads.
- the SAN switches and the V7000 have to be discovered and successfully accessed
- the system that you will capture has to have the Common Agent discovered and has to have the
activation engine code installed and the activation engine has to be configured. The
activation engine code resides on the FSM and can be copied using scp after which it has ot
be installed on the host using specific java code:
opt/ibm/director/proddata/activation-engine/vmc.vsae.tar
root@1.1.1.1:/tmp/
export JAVA_HOME=/usr/java5/jre
untar and execute the AE install script
After installing the activation engine and before trying to capture make sure you run an inventory
collection on the FSM, Farm, VIOS, host system.
76
After you customize the OS that you will capture you have to run the AE.sh script:
/opt/ibm/ae/AE.sh --reset
This will prepare the OS for capture and will also shut down the virtual server. After the system is
shut down you can initiate the capture by right clicking the Virtual Server element in resource
explorer or by following the VMControl links.
In case you want to make some changes and capture the server again you have to reset the
activation engine by issuing these commands:
rm /opt/ibm/ae/AP/*
cp /opt/ibm/ae/AS/vmc-network-restore/resetenv /opt/ibm/ae/AP/ovfenv.xml
If you have done everything correctly, which I know you did, you will now have your customized
image ready to deploy. Please note that the CPU, Memory, Network, Storage information is all
captured form the original image, all new deployments will use this information as default,
which of course you can change later.
77
2011
5.3
The FSM comes with a very useful feature called Electronic Service Agent, also known as call
home. I highly recommend enabling this, it will automatically create a service call with IBM
Support in case of any hardware failures, it will also collect and upload the logs.You do need to
have an active support agreement contract and you have to provide the customer number. It's
also a good idea to check on the IBM Support website that the correct product is registered.
79
Part of the advanced RAS features of the FSM include predictive hardware alerting which can be
enabled by creating an Event Automation Plan that will filter out from the event log the specific
alerts and will log them in the error log. This plan can also include an action to send and email
or integrate into IBM Tivoli Monitoring, for example.
80
Go to the Plugins tab, Automation Manager, Event Automation Plans, create new:
81
Select All Systems as target, this will ensure that all predictive hardware alerts are captured from
all systems:
82
Change to Advanced Event Filters and select Hardware Predictive Failure Alerts:
83
Next step is to select what action to do in case an alert is raised. If you don't see an appropriate
action in the list click create:
84
Select from the list send email, this way the recipient will be informed:
85
Note you can only use one address for action, and you can customize what to include in the
message using variables:
86
Next step is to finish the creation and activate the plan. Since this is related to predictive
hardware alerts it will be active 24/7, with other plans you have the flexibility to chose the
timeframe. For example if your service agreement is only from 9 to 5 you on weekdays you
can set the plan to be active only during this timeframe.
Another useful feature is monitoring, which can include filesystem usage, CPU, memory, process
monitors etc.
For each monitor you can define a threshold with a severity and you can create automation plans
for these as well.
87
2011
FCoE
5.3
FCoE
IBM Power Systems Technical University Athens 2013
Support for FCoE in Power nodes comes with the new 8 port, dual asic Converged Ethernet
Adapter and the CN4093 10Gb converged switch. Due to current limitations of the available
CN4093 switch only 6 of these can actually be used and if you plan to split the adapter for 2
VIO servers you will only be able to use 2 ports for each.
When used in converged mode each physical port is split into 2 providing Fibre Channel and
Ethernet traffic.
The CN4093 converged switch can act as the primary SAN director when running in full fabric
mode or as an NPV gateway to provide pass-through traffic to a top of rack SAN director. It
does use a Qlogic fabric thus full fabric integration with Brocade is not possible, in this case I
recommend using NPV. It has 12 omniports that can be configured in pairs only, and you can
attach a SAN SFP+ for direct connection of storage systems or SAN directors.
89
FCoE
IBM Power Systems Technical University Athens 2013
90
FCoE
IBM Power Systems Technical University Athens 2013
You have to define a new vlan for FCoE traffic only, the accepted industry default is 1002, and
configure it accordingly:
#vlan 1002
#member INTA2-INTA14,EXT6-EXT7
You have to include all host ports and the omniports configured as FC in this vlan. Then if you
plan to use the switch in pass through (NPV) mode:
#npv enable
#npv traffic-map external-interface EXT6-EXT7 (providing the FC ports that will be connected
to your SAN Director switch)
Note that it is not supported to transport vlan 1002 through an ISL link.
91
FCoE
IBM Power Systems Technical University Athens 2013
92
FCoE
IBM Power Systems Technical University Athens 2013
93
94
95
ibmtechu.com/gr
IBM Power Systems Technical University Athens 2013
Win prizes by
submitting
evaluations online.
The more evalutions
submitted, the
greater chance of
winning
96
ibm.com/training
Now is the time to explore your options for
additional training. One stop shopping for all
your technical training needs.
1.
1.
ibm.com/training/trainingpaths
IBM Training Paths These flowcharts map
out the sequence of classes you need, to
obtain a specific skill or professional
certification. Get started today!
ibm.com/certify
Stand out from the crowd when you earn a
valuable credential of an IBM Certified
Professional. Find certifications by product or
solution.
97