????????????????????????????????????????????????????????????????????Chris Churchey – Principal ATS Group, LLC [email protected] (610-574-0207) October 2013
Exploit PowerVM Features to Maximize Performance & Efficiencies
© 2013 IBM Corporation
????????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
2
© 2013 IBM Corporation
Exploit PowerVM Features to Maximize
Performance & Efficiencies
Multiple Shared Processor Pools (MSPP)
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
What is MSPP?
? Group of physical processors
? Shared among multiple logical partitions
? The sum of the assigned capacity for each LPAR cannot exceed the total capacity of the processor group
? Applies to Power6 and above
3 © 2013 IBM Corporation
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
Why use MSPPs?
? Reduce software license (e.g. # of cores) costs by placing LPAR(s) in their own pool – Customers have reported saving $100K to >$8M a year
? Share the software license across LPAR(s) by placing them in the same pool (e.g. Float across Prod & Dev)
? Limit a group of LPAR(s) to a combined MAX CPU consumption to protect others
– Example: Power 770 with 32 cores
• Dev LPAR(s) pool Max CPU=5
• Prod LPAR(s) pool Max CPU=32 (All)
4 © 2013 IBM Corporation
???????Exploit PowerVM Features to Maximize Performance & Efficiencies
MSPP Analysis
1. Identify candidate LPAR(s)
– Software licensed by # of cores currently Dedicated or Capped, or license ALL cores on
the server
2. Collect physical CPU usage for candidate LPAR(s)
3. View the candidate LPAR(s) over a period of time
4. Determine usage mix and observed Peaks
– The pool’s Maximum Processing Units = “peak observed + projected growth + buffer”
? NOTE: Galileo Performance Explorer makes this easy using Virtual Groups to view candidate LPAR(s)!
5 © 2013 IBM Corporation
???????????Exploit PowerVM Features to Maximize Performance & Efficiencies
MSPP Analysis
6 © 2013 IBM Corporation
p795 with 23 LPARs, peaking 23 cores
???????????Exploit PowerVM Features to Maximize Performance & Efficiencies
MSPP Analysis
7 © 2013 IBM Corporation
Selected 7 Oracle LPARs, peaking 12 Cores
???????????Exploit PowerVM Features to Maximize Performance & Efficiencies
MSPP Analysis
8 © 2013 IBM Corporation
“What If” Shared Pool Analysis
???????????Exploit PowerVM Features to Maximize Performance & Efficiencies
MSPP Analysis
9 © 2013 IBM Corporation
Performance View of Pool 0 and Pool 1
???????????Exploit PowerVM Features to Maximize Performance & Efficiencies
MSPP Analysis
10 © 2013 IBM Corporation
Detailed Performance View of Pool 1 (Oracle Pool)
???????Exploit PowerVM Features to Maximize Performance & Efficiencies
MSPP Resources
? IBM PowerVM Virtualization Introduction and Configuration www.redbooks.ibm.com/abstracts/sg247940.html
? Exploiting IBM PowerVM Virtualization Features with IBM Cognos 8 Business Intelligence www.redbooks.ibm.com/abstracts/sg247842.html?Open
? Try it! MSPP Pools can be dynamically created, move LPAR(s) in & out, and increase/decrease MAX
? ATS can help analyze and implement MSPP
11 © 2013 IBM Corporation
????????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
Exploit PowerVM Features to Maximize
Performance & Efficiencies
Shared Ethernet Adapters (SEA)
12 © 2013 IBM Corporation
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
What is a SEA?
? Group Virtual I/O Server component that bridges a physical Ethernet adapter and Ethernet adapters
? LPAR(s) on the virtual network can share access to the physical network
? LPAR(s) on the virtual network can communicate with stand-alone servers and logical
partitions on other systems
? Eliminates the need for each client logical partition to a dedicated physical adapter
13 © 2013 IBM Corporation
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
Why use SEAs?
? Required for Live Partition Mobility (LPM)
? Eliminate dedicated adapters to LPARs
? Reduce network switches, ports and cabling
? Reduce # of adapters, thus reduce slots/IO-drawers
? Share network bandwidth across LPARs
? Balance CPU load across VIO’s (1 often 100% Busy)
? Maximize network throughput using ALL adapters
14 © 2013 IBM Corporation
?????????Exploit PowerVM Features to Maximize Performance & Efficiencies
SEA Options
? SEA-NIB (Network Interface Backup) + vSwitches
– (2) Virtual Ethernet adapters per LPAR
– EtherChannel used on Client LPAR to designate Primary VIO – Script to balance Even/Odd LPAR to primary Even/Odd VIO – Preferred for Max Network Throughput and Availability
– Multiple VLANs NOT required
? SEA-HA (High Availability)
– (1) Virtual Ethernet adapter per LPAR
– All LPAR(s) network goes through 1 VIO w/failover to 2nd VIO – 2nd VIO Ethernet ports idle, unused performance
? SEA-HA Load Sharing (aka Load Balancing)
– Requires multiple VLANs which are balanced between VIO’s – It chooses which VLANs go to which VIO’s
– Requires manual ‘chdev’ to restart if failover
15 © 2013 IBM Corporation
???????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
SEA Architecture – NIB w/vSwitches
16 © 2013 IBM Corporation
1Gb (80-100MB/s) + 1Gb (80-100MB/s) Load Balance=200MB/s
????????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
SEA Architecture – HA
17 © 2013 IBM Corporation
1Gb (80-100MB/s) 1Gb (80-100MB/s) Backup-Wasted
??????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
SEA Architecture – HA Load Sharing
18 © 2013 IBM Corporation
1Gb (80-100MB/s) + 1Gb (80-100MB/s) Load Sharing=200MB/s
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
SEA Analysis
1.
Collect physical CPU and SEA usage for VIO LPARs
2.
Look for:
3. 4.
If using SEA-NIB today, balance script on all LPARs
19
© 2013 IBM Corporation
– View VIO LPARs over a period of time
– 1 VIO with high/max CPU and other minimal
– 1 VIO SEA at max throughput (flat line) and other minimal
If using SEA-HA today, use Load Sharing feature
– NOTE: Galileo Performance Explorer can display this performance information quickly
and easily!
??????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
SEA Analysis – NIB
? Host SEA Network Packet Send & Receive ? Host Physical CPU Utilization Summary Throughput
20 © 2013 IBM Corporation
??????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
SEA Analysis – HA
? Host SEA Network Packet Send & Receive ? Host Physical CPU Utilization Summary Throughput
21 © 2013 IBM Corporation
????????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
Exploit PowerVM Features to Maximize
Performance & Efficiencies
NPIV vs. vSCSI Performance
22 © 2013 IBM Corporation
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
What is NPIV and vSCSI?
23
© 2013 IBM Corporation
NPIV
vSCSI
? Allows a single HBA to register multiple WWPNs
? A SCSI disk, or a volume or file in a VIO Server that is exported to a virtual IO client
? Each virtual server presents a different WWPN to the SAN
? LUNs implement a sub-set of the SCSI protocol
? Each virtual server will see its own unique storage
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
Why NPIV over vSCSI?
? Higher throughput from client LPAR
? Higher IOps from client LPAR
? Better hdisk Service Time from client LPAR
? Less CPU usage by the VIO LPAR
? Load Balance IO across HBAs from client LPAR
24 © 2013 IBM Corporation
?????????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
NPIV vs. vSCSI Architecture
25 © 2013 IBM Corporation
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
NPIV/vSCSI Test configuration
? AIX LPAR (gvicaix71-10) on a p700 blade and IVM – (2) 8Gb HBA to V7000
• (4) 100GB vdisks (LUNs) created…same mdisk/array group
– (vSCSI) =?(2) 100GB LUNs mapped to VIO, mkvdev to AIX LPAR
• hdisk3 and hdisk4 on gvicaix71-10
– (vfc) =?(2) 100GB LUNs mapped to (2) wwpn’s on AIX LPAR
• hdisk1 and hdisk2 on gvicaix71-10
? ndisk64 Seq./Random Read/Write 4K/256K tests
26 © 2013 IBM Corporation
????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
NPIV/vSCSI Test configuration
27 © 2013 IBM Corporation
??????????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
NPIV vs. vSCSI IOps
28 © 2013 IBM Corporation
NPIV vSCSI
NPIV: Over 500 more IOps!
??????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
NPIV vs. vSCSI Read Service-Times
29 © 2013 IBM Corporation
NPIV
vSCSI
NPIV: 0.75ms faster!
????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
NPIV vs. vSCSI Write Service-Times
30 © 2013 IBM Corporation
NPIV
vSCSI
NPIV: 0.50ms faster!
??????Exploit PowerVM Features to Maximize Performance & Efficiencies
31 © 2013 IBM Corporation
Questions and Answers
???????Exploit PowerVM Features to Maximize Performance & Efficiencies
? We can help analyze and implement. Contact us! ? Check-out Galileo Performance ExplorerTM
– Visit our booth for a hands-on demo
– Complimentary* no-strings attached 6 months use for Conference attendees – Budget-friendly subscription pricing request
* First time Galileo user
32 © 2013 IBM Corporation
www.GalileoSuite.com
??????????????????Exploit PowerVM Features to Maximize Performance & Efficiencies
Referenced Material
?IBM PowerVM Virtualization Introduction and Configuration www.redbooks.ibm.com/abstracts/sg247940.html
www.redbooks.ibm.com/abstracts/sg247842.html (Managing and Monitoring) ?SEA?NIB vs. SEA?HA vs. SEA?HA?Load Sharing
www.ibm.com/developerworks/mydeveloperworks/blogs/aixpert/entry/shared_ethernet_adapter_sea_failover_with_l oad_balancing198?lang=en
www.ibm.com/developerworks/mydeveloperworks/blogs/AIXDownUnder/entry/shared_ethernet_adapter_load_sharin g_and_virtual_switches133?lang=en
? SEA?NIB?vSwitches
www? 03.ibm.com/support/techdocs/atsmastr.nsf/fe582a1e48331b5585256de50062ae1c/81c729a840b213b98625779e000722f4/$FI LE/PowerVM?VirtualSwitches?091010.pdf
ibmsystemsmag.blogs.com/aixchange/2011/09/higher?availability?for?vio?clients?an?alternative.html
? NPIV www.ibm.com/developerworks/aix/library/au?NPIV/index.html
?Galileo Performance Explorer www.galileosuite.com
* First time Galileo user
33 © 2013 IBM Corporation
Categories: News