vmug belgium - the devastating impacts of virtual sprawl
TRANSCRIPT
Who is Chris Wahl? • VMware Certified Design Expert (VCDX) #104 ▫ Data Center and Network Virtualization
• Tech, Lab, Certs, Reviews, Videos ▫ WahlNetwork.com
• CCNA Data Center & vCloud Courses ▫ Pluralsight.com
• Virtualization User Podcast as a Service ▫ VUPaaS.com
Networking for VMware Administrators Paperback, Kindle, eBook, EPUB, MOBI, PDF 100% of my profits go to Alzheimer’s Association
Social Media
VirtualSprawl (vûr′cho ̅o ̅-əl sprôl) n. A virtual environment with insufficient consumption control and poor lifecycle management
Questionable Reclaiming
Methods
The “Wild West” You can’t have it both ways Rapid provisioning requires rapid reclamation
Projects no longer own hardware.
They lease a logical slice of resources
Monster VM sprawl • Fancy term for a big VM • Software and hardware advancements • Additional considerations
Common misunderstandings • Beware of “always” ▫ Adding CPUs makes a VM faster ▫ Adding Memory makes a VM faster
• Abuse of virtual resources
Identifying performance issues Find out who is ruining the party
Overallocation • “Too much of a good thing” ▫ Right-sizing and waste-finding exercises ▫ vCPU is a common culprit ▫ 4 vCores : 1 pCores is a common goal
Squeeze the juice!
Common warning signs • High %RUN and %IDLE variations • Find %RDY or %CSTP • > 4 vCPUs per pCore
Common warning signs • Host Swap Rate > 0 • Ballooning Driver Frequently Activates
Sub-Optimal Hardware Allocation Larger VMs have needs
Sub-Optimal Hardware Allocation • Sprawl across hardware ▫ Common with older hardware
• NUMA ▫ Check NHN & N&L ▫ Is vNUMA enabled? ▫ > 8 vCPUs? ▫ vProc & vCore match pProc and pCore?
The Michigan Left
NUMA in a nutshell
• Yellow = NUMA Nodes • Pink = NUMA Home Node • Blue = NUMA Locality %
Optimal Hardware Allocation
Sub-Optimal Hardware Allocation
NUMA Node 0 NUMA Node 1
Poor Locality
Competition for resources Those annoying noisy neighbors
DRS is OK, but … • It’s not a silver bullet • Tends to worry over CPU and RAM ▫ … while being ignorant of the larger picture
• A few critical metrics from the field ▫ Licensing configuration ▫ %RDY and %CSTP ▫ Bus device utilization (HBAs, NICs) ▫ Logical placement
3rd Party Tools • Focused more on the tenants as services ▫ Building licenses stacks (SQL, Oracle, etc.) ▫ End-to-end data path ▫ Advanced features
• A few examples ▫ VMTurbo, Cirba, Proactive DRS (fling)
Taking charge! Education is key
You should be the trusted advisor • It’s not the vendors or partners ▫ You’re the one left on the hook, later
• Perform data collection ▫ Which VMs are priority? ▫ How bad is it? ▫ Use free tools (vCOps foundation version)
Build the big win • Find the worst offender • Collect baseline performance • Make change recommendations • Implement the change ▫ Use change control
• Re-baseline performance • Hold a delta review
Set Policies and Precedence Provisioning Requests
Sit in as the trusted advisor Communicate with the application owner(s) Take ownership Have documented hard and soft limits Exceptions should be rare and well understood
Example Exemptions Most common: VBCA
Tier 1 SQL, Oracle, Exchange, etc. Silo into VBCA specific cluster?
New & unknown applications Start small, work upwards
Thank you for a great session! For more goodies, check out these resources:
Visit WahlNetwork.com for Deep Dives, Home Lab Tips, Reviews, and More!
Check out “Networking for VMware Administrators” by VMware Press
Over 25 hours of my video training content is on Pluralsight.com
Nearly 100 free HD videos on the WahlNetwork YouTube Channel