Presentation #2 - Individual CMG Regions and SIGs

Report
Metron
vSphere vs. Hyper-V
Performance Showdown
Objectives
• Architecture
• Available metrics
• Challenges in virtual environments
• Test environment and methods
• Results, conclusions, and caveats
vSphere Architecture
Small hypervisor footprint
Hyper-V Architecture
Windows OS required - Larger footprint
vSphere Memory Management Features
Transparent page sharing
Memory borrowing
Memory compression
…and Paging
Hyper-V Memory Management Feature
Dynamic memory for enlightened Windows VMs
Key Performance Metrics
vSphere
Hyper-V
CPU
Avg. CPU Usage in MHz
CPU Ready Time
CPU
Hypervisor Logical Processor %
Hypervisor Virtual Processor %
Memory
Avg. Memory Usage in KB
Balloon KB
Swap Used KB
Consumed
Active
Memory
Dynamic Memory Balancer
Dynamic Memory Pressure
VM Vid Partition
I/O
Queue Latency
Kernel Latency
Device Latency
I/O
Virtual Storage Device
Virtual Network Adapter
Challenges in Virtual Environments
- Clock skew
- NUMA (non-uniform memory access) scheduling
- Pass-through I/O
- Workload definition
- 32 versus 64 bit OS and applications
- Where to use solid state I/O devices
- Storage tiers
- Hidden overhead
- etc...
Test Environment
-
AMD Phenom II 3.3 GHz
8 GB RAM
1TB Hitachi 7200 RPM HD SATA 2 interface
1GB Onboard network interface
- vSphere 5
- Hyper-V role installed on Windows 2008 R2 SP1
- 2 x Windows 7 SP1 VM with integration services
- 2 x CentOS 6.2 VM with integration services v3.2
- Simple custom benchmarks using ActiveState perl v5.14
- cpu.pl, disk.pl, mem.pl, net.pl
Testing Methods
Individual Tests
Hyper-V Hostname
Hyper30
Virtual Machines
win_25, win_26
centos_11, centos_12
Win7 VMs configured
with 512MB min, 64GB
max dynamic memory.
Centos VMs 2GB.
VMs _25, _12
configured with 2 vCPU
vSphere Hostname
192.168.0.99
* No Pass-Through
- CPU w/1 process
- CPU w/2 process
- Disk on 1vCPU VMs
- Mem on 1vCPU VMs
- Net on 1vCPU VMs
Combined Tests
All VMs running CPU
and Mem. 2 vCPU VMs
running Disk and Net.
Each set of testing was
run identically on the
same host using both
hypervisors
Results - Individual VM CPU
Surprisingly Win7 on Hyper-V appears slower
Results - Individual VM CPU
Extra vCPU no help with one process
Results - Individual VM CPU
Twice the work in the same time
Results - Individual VM CPU
Summary of CPU results
- Windows CPU performance on Hyper-V was significantly slower
- Two vCPUs running a single process had little negative impact
Results - Individual VM Disk
Write a 512 MB file
Results - Individual VM Disk
Read the 512 MB file
Results - Individual VM Disk
Abysmal performance for Windows on Hyper-V
Results - Individual VM Disk
Summary of disk I/O results
- Random I/O on a Hyper-V dynamic disk had terrible performance
Results - Individual VM Memory
No penalty for dynamic memory
Results - Individual VM Network
Slight advantage for vSphere
Results - Combined Test
And now… for the grand finale
All workloads running at the same time on multiple VMs
The winner is….
Results - Combined Test
Draw - except disk workload on Hyper-V did not finish
CPU Details - Both platforms on the same chart
Detail CPU Metrics
- vSphere VM ready time
- Hyper-V Guest run time
CPU Details - vSphere CPU Ready Time
CPU Details - Hyper-V Guest Run Time
Detail Memory Metrics
- vSphere memory consumed by VMs
- vSphere memory ballooning
- vSphere paging
- Hyper-V memory balancer average pressure
- Hyper-V memory current pressure
- Hyper-V physical page allocation
Memory Details - vSphere Ballooning
Memory Details - vSphere Paging
Memory Details - vSphere Consumed
Memory Details - Hyper-V Memory Balancer Pressure
Memory Details - Hyper-V Memory Current Pressure
Memory Details - Hyper-V Pages Allocated
Detail I/O Metrics
- vSphere queue latency
- vSphere device latency
- Hyper-V disk throughput
I/O Details - vSphere Queue Latency
I/O Details - vSphere Device Latency
I/O Details - Hyper-V Disk Throughput
Interesting Observations
- Hyper-V dynamic memory
- Hyper-V dynamic disk device performance
Dynamic Memory in action - High Pressure
Dynamic Memory in action - Low Pressure
Hyper-V - Random I/O on Dynamic Disks
Chart from Hyper-V MSDN Blog by: Tony Voellm
Conclusions, Caveats, and Final Thoughts
Overall the combined results were surprising close
Individual tests produced some interesting findings
- Windows CPU performance on Hyper-V was significantly slower
- Two vCPUs running a single process had little negative impact
- Random I/O on a Hyper-V dynamic disk had terrible performance
- Hyper-V dynamic memory worked great with no performance penalty
Caveats
- Workloads were very general and dependent on perl implementation
- Many more variables could be taken into account
- Result may be different on other hardware
Running benchmarks in your own environment should be
done to help make the best informed decisions.
Thank you for attending
vSphere vs. Hyper-V
Charles Johnson
Metron-Athene Inc.
[email protected]

similar documents