VMworld 2013
Sheldon Brown, SRP
Girish Manmadkar, VMware
Learn more about VMworld and register at http://www.vmworld.com/index.jspa?src=socmed-vmworld-slideshare
Dev Dives: Streamline document processing with UiPath Studio Web
VMworld 2013: How SRP Delivers More Than Power to Their Customers
1. How SRP Delivers More Than Power to
Their Customers
Sheldon Brown, SRP
Girish Manmadkar, VMware
VAPP5105
#VAPP5105
2. 2
Agenda
Introduction
Before We Start - Homework
Figuring It All Out
What It All Looks Like Now
After the Dust Settled
Summary
Questions
4. 4
Girish Manmadkar – Consulting Architect
With VMware since 2007 in various roles, focusing on providing
VMware solutions around Business Critical Application with major
focus around SAP
More than 22+ years experience starting with SAP R4.6
Regular speaker at VMworld since 2007
Recent VMware Partner Exchange, EMCWorld events
5. 5
Sheldon Brown – Manager SAP Technology
Responsible for SAP Technology (BASIS,Development/Integration)
21 years with SRP
10 years with Server and Storage Team
2 years with SAP implementation team
9 of 10 years at VMWorld
sheldon.brown@srpnet.com
6. 6
Who We Are – Salt River Project Agricultural and Power District (SRP)
Power and Water supplier to the 1,000,000+ customers in the
Phoenix metro area
Third largest public power company in the country
100+ years old
5000+ employees
Slow moving, risk adverse
70% virtualized
http://www.srpnet.com
8. 8
Best Practices
Use the latest processor generations, especially with EPT (Intel
XEON 55xx, 65xx, 75xx and newer) or RVI (AMD Opteron 83xx,
84xx, 61xx and newer) which bring performance benefits
Assigning vCPUs to virtual machines
• ESXi 5.1: shows scaling for virtual machines > 8 vCPUs
• Understand the v/NUMA architecture for larger VMs
You can overcommit CPUs, but reasonably
• Start with no CPU overcommitment. When the average utilization of SAP
instances and the average idle time on the host are known, accordingly raise
the number of vCPUs on the host (for example, adding virtual machines). Use
VMware vSphere Distributed Resource Scheduler (DRS) capabilities to control
and balance overall workload.
• If performance problems are seen, first reserve CPU to see if the performance
problems are a result of resource contention
9. 9
Best Practices (cont.)
• Do not overcommit memory. This is not recommended at all with SAP.
Reserve 100% of the assigned memory.
• Installation of VMware Tools is mandatory to avoid time conflicts and have all
necessary drivers (for example, new network cards)
• Timekeeping: Use NTP (Network Time Protocol) on the ESX host and guest
as per Timekeeping best practices for Windows, including NTP and
SAP Note 989963
http://kb.vmware.com/kb/1318
• Use multiple virtual SCSI controllers for the database virtual machine and
virtual machines with high I/O load. The use of multiple virtual SCSI controllers
allows the execution of several parallel I/O operations inside the guest OS.
• SAP requires the implementation of the new virtualization aware monitoring.
See SAP Note 1409604.
• When using VMware snapshots, follow Best practices for virtual machine
snapshots in the VMware environment. Make sure that the virtual machine has
no active snapshot before reporting performance problems.
http://kb.vmware.com/kb/1025279
10. 10
Best Practices for Linux
Disabling the Linux I/O scheduler
• Set kernel parameter "elevator=noop"
• Red Hat KB 5428 (applies to RHEL and SLES)
Choose the optimal SAP memory model
• MAP memory model (SAP Note 386605) for CPUs without EPT/RVI
• STD memory model (SAP Note 941735) for CPUs with EPT/RVI
11. 11
vSphere – ESX CPU Scheduling
Determine size of hardware NUMA nodes on the server
Size the virtual machine as small as possible
• Too many vCPUs creates scheduling overhead
• Too much RAM results in accessing RAM from a far NUMA node
Enable simultaneous multithreading
• More hardware execution contexts are available
Generally CPU affinity not required
12. 12
Best Practices for Databases
Oracle
• SAP Note 1173954 states support with minor restrictions
• Oracle on VMware vSphere Essential Database Deployment Tips
http://www.vmware.com/resources/techresources/10101
• VMware Alliances statement
http://www.vmware.com/solutions/partners/alliances/oracle-vmware-
support.html
13. 13
Best Practices for Databases – I/O Subsystem
Store data files and log files on separate physical devices and distribute
data files as per SAN Volume Controller as defined by the storage
vendor’s best practices
Quote from SAP on SQL Server best practices guide
• “The number of disks is determined by the throughput in I/O operations per
seconds (IOPS). Therefore discussions around storage investments or
configurations should be driven by IOPS”
• http://www.sdn.sap.com/irj/scn/index?rid=/library/uuid/4ab89e84-0d01-0010-
cda2-82ddc3548c65
IOPS used by storage and OEM vendors to size SAP storage on
VMware – Similar to physical
The standard SAP rules for databases apply!
• SAP Note 806554 for general I/O intensive DB tasks
• SAP Note 592393 and 793113 for Oracle
14. 15
Best Practices for vStorage APIs (cont.)
VMware vSphere® Storage APIs – Array Integration (VAAI)
• Has nothing to do with backup – it is a better way of integrating storage
capabilities into vSphere
• vStorage APIs for Array Integration FAQ
http://kb.vmware.com/kb/1021976
• Follows the vSphere lifecycle
• Offloads specific storage operations to compliant storage hardware
• Write Same/Zero
• Eliminating redundant and repetitive write commands
• Fast/Full Copy
• Leveraging array ability to mass copy, snapshot, and move blocks
• Hardware Offloaded Locking (ATS)
• Stop locking LUNs and start locking only blocks
15. 16
Best Practices for Network
Standard vSphere networking guidelines apply
• Separate infrastructure traffic from virtual machine traffic: use physical NICs or VLANs
• Use NIC teaming for availability and load balancing (group physical NICs connected to
the same physical network)
• Take advantage of network I/O controls to converge network and storage I/O onto
10GbE
• Take advantage of the dvSwitch and/or Cisco Nexus 1000v to simplfy the architecture
and support
Use VMXNET3 adapter in guest OS
• Improved performance
• Helps with network traffic in three-tier setup between application and DB VM
• SAP Batch Job Performance on vSphere
http://blogs.vmware.com/performance/2010/02/sap-batch-job-performance-on-
vsphere.html
16. 17
Best Practices – Further Information
SDN Forum SAP on VMware
• http://forums.sdn.sap.com/forum.jspa?forumID=471
• Most complete link collection around SAP on VMware
SAP Notes
• Following the SAP Notes is mandatory!
VMworld Sessions
• Various documents – from high level overview to technical deep dive
Performance Tests
• Continuous performance measurement for databases and SAP software
19. 20
Figuring It All Out – the Team
Team Effort
Consultants
• Accenture
• SAP
• HP
• VMware
• NetApp
• Cisco
SRP Employees
• Server
• Storage
• Network
• BASIS
• Strategy and Planning
• Executive Management
20. 21
Figuring It All Out – What we needed to decide
Datacenters
Host Servers
Virtual Servers
Storage
Network
Database
Operating System
Infrastructure Security
21. 22
Figuring It All Out – SAP Environment Description
ECC, SRM, PI, BW, SolMan, BOBJ, GRC, Portal, Data Services
Bolt Ons
• Vertex
• OpenText
• BPC
Landscapes (Production, Quality, Test, Development, Sandbox, Training)
• Production
• Quality
• Test
• Development
• Sandbox (Technical and Functional)
• Training
5000+ users
850+ unique per day
22. 23
Figuring It All Out – Quick Sizer Results (nothing “quick about it”)
User Based SAPs
Memory and Disk Sizing
24. 25
What It All Looks Like Now – Datacenters and Network
Two datacenters
• 30 miles by the wire apart
• Connected by 20Gbps (Layer 2) 16 Gbps (Layer 3)
Network
• Cisco 7k series core
• Cisco 5k series edge
• 10Gbps network
25. 26
What It All Looks Like Now – Physical Servers
24 ESX Hosts (2 clusters of 12 at each data center)
• HP DL580 G7
• vSphere 5.1 U1
• 4 proc 10 core
• 512 GB RAM
• 2 10GBps ports
6 ESX Host Stretch Cluster (3 at each data center)
• HP DL580 G7
• vSphere 5.1 U1
• 4 proc 10 core
• 256GB RAM
• 2 10GBps ports
26. 27
What It All Looks Like Now – Virtual Servers
Production – 76 servers
Quality – 59 servers
Test – 27 servers
Development – 32 servers
Sandbox (Technical and Functional) – 40 servers
Training – 3 servers
Administration – 3 servers
Total – 240
Operating Systems – Red Hat Enterprise Linux (RHEL) 5.x to 6.2
and Windows 2008 R2
27. 28
What It All Looks Like Now – Storage
4 NetApp 6280’s (2 at each data center)
• NetApp Data ONTAP Release 8.1.2P3 7-Mode
• Theoretical capacity 77,000 IOPS (440 disk * 175 IOPS)
• Highest recorded capacity 8212 IOPS
• Raw Capacity:154,493GB or 150TB
• Consumed Capacity: 61519GB or 60TB
• Extensive use of Snap Shot Manager for SAP
28. 29
What It All Looks Like Now – Network Architecture
Datacenter 2Datacenter 1
29. 30
What It All Looks Like Now – SAP vSphere Host Configuration
Datacenter 1 Datacenter 2
31. 32
What It All Looks Like Now – Monitoring
SAP Solution Manager – Application
Quest vFoglight – Virtual Server Performance
Top, Satellite – Operating System
CA Wily Interscope – Java
NetApp ONCommand Unified Manager – Storage
HP SIM – Server Hardware
Q1, Cascade Shark, eHealth – Network
CA Spectrum – Operations
32. 33
What It All Looks Like Now – Database
Oracle 11g (Corporate Standard)
DataGuard for replication
One virtual server per database
Unlimited Oracle for SAP
33. 34
What It All Looks Like Now – Disaster Recovery/High Availability
Completed a Business Impact Analysis
12 hour RTO
Close to zero RPO
Load-balanced dialog instances
Central Messaging instance at primary datacenter
Database instance at primary datacenter
Database replication happening real time
34. 35
Datacenter 1
What It All Looks Like Now – Disaster Recovery/High Availability
Replication
Log Shipping
Datacenter 1 Datacenter 2
35. 36
What It All Looks Like Now – Why We Virtualized SAP on VMware
Cost Effective
Flexibility
Higher Availability
Faster Deployment
Easier to recover in disaster
Higher Server Density = Lower Data Center Costs
Virtualization is the SRP corporate standard
VMware is the leader
Favorable license agreement with SAP
36. 37
What It All Looks Like Now – Refresh Methodology
Prepare the source and target systems
Snap Clone production using Snapshot Manager for SAP
• Mount cloned copies to targeted servers
• Changes the SID
• Minutes instead of hours
Post clone cleanup
Validate target systems
Scrub production data
Enjoy
38. 39
After the Dust Settled - Lessons Learned
Overbuild
Plan for disaster now
Plan out storage configuration early
Pay attention to security early
• Infrastructure
• OS and Application hardening
39. 40
After the Dust Settled - Successes
Flexibility
Speed in reaction
Environment Refresh rate
Spin up environments quickly
Saved $1,000,000 in hardware by virtualizing Oracle
40. 41
After the Dust Settled - Things to Come
Release 2
• Work Management?
• HR
• More Supply Chain
• More Finance
• Customer Information?
Mobility?
HANA?
41. 42
Summary
Now that we are done – don’t forget your home work
Figure it all out
What It All Looks Like Now
After the Dust Settled