How to Resolve Bottlenecks and Optimize your Virtual

Download Report

Transcript How to Resolve Bottlenecks and Optimize your Virtual

How to Resolve Bottlenecks
and
Optimize your Virtual Environment
Chris Chesley, Sr. Systems Engineer
[email protected]
Today’s Objectives
 Overview of the problems often encountered in a
virtual environment.
 How to identify current performance bottlenecks
 How to find capacity in your environment
 How to optimize your virtual machines
 How to identify wasted storage
New vs. Old Data Center
Pre-Virtualized Data Center
• One application – One server
• No sharing of memory and CPU
• Limited sharing of storage
Virtualized Datacenter
• Shared memory and CPU
• Massively shared storage
• Performance = capacity
• Vmotion/DRS – Dynamic nature
• Change from known to unknown state when VM added or changed
Takeaway
If not closely monitored and managed,
sharing of resources will lead to
performance problems and downtime.
Resource Utilization
What’s shared
 Physical CPU / Cores (not sockets)
 Physical Memory
 Physical Disk
 Storage (how many GB am I using)
 I/O (reads, writes, disk latency, disk queue latency)
 Physical Network
Boundaries in Virtual Environments
Capacity issues occur at any level or resource type
 VMs
 Host
 Cluster
 Resource Pools (Vmware only)
 Data Center
How to Identify Current Capacity Bottlenecks
 Out of the box Management consoles: vCenter client (Vmware) and Hyper-V
Manager or Virtual Machine Manager (Hyper-V):Real time per-host stats, percluster, and per-VM statistics
 Esxtop (Vmware) or PerfMon (Hyper-V): per-host statistics
 VIM API and SDK: allows software to collect only the statistics they want
Performance Monitoring Options
 For every Cluster, Resource Pool, Host, and VM, using
VirtualCenter to examine memory, storage, CPU and network
utilization over a period of at least 1 week
 Very time consuming process
– (Clusters + Resource Pools + Hosts + VMs) X 5 resources (CPU, storage,
RAM, network, disk i/o) = # of charts to review
– e.g. (3 Clusters +3 RPs+ 50 Hosts + 500 VMs) X 5 = 2,780 charts to examine
 Requires ongoing attention: at least several times a week
Identifying Available Resource Capacity
Select a Cluster, Resource Pool, or a Host
Get info on available memory, storage, CPU, disk i/o, and
network i/o
Calculate an average VM footprint
 Memory, CPU, storage, disk i/o
Where do I put new VMs?
Identifying Available Resource Capacity
Apply an average VM footprint to every resource type to
see which resource you will run out of first
That’s how many more VMs you can fit into Hosts,
Clusters, or Resource Pools
Where do I place new VMs
Data Center view – Active/Active
 DR Data Centers may have capacity
Predicting Future Capacity Bottlenecks
Model additions of new VMs
 Understand current utilization on all resource
types
 Make necessary changes to compensate for
current and future growth
Implement iron clad change control process
 Maintenance window and workload
requirements
 Cluster failover configuration
 Resource Pool configuration
 Powered down VMs
Optimize your virtual machines
• Very easy to create VMs, not easy to know how many
resources to give them.
• No automated clean up
• According to the Information Technology
Infrastructure Library (ITIL), capacity management is
the discipline that ensures IT infrastructure is used in
the most efficient, predictable and cost-effective
manner.
• Goal of capacity management: Finding the balance
between density and performance.
Defining which Metrics to Monitor
 Allocated vs. consumed resources.
 CPU: The important metric to measure is CPU utilization.
 Memory: Use memory consumed in most cases when evaluating
memory utilization.
 Memory consumed measures how much memory each VM is using on the
physical host minus an memory that is shared by other VMs.
 Storage: The best way to monitor storage is to look at each vmdk file
from the guest OS perspective and look at utilization.
Identifying Allocated Resources
Allocated resources, limits and reservation info can be easily
collected from vCenter or a 3rd party solution.
Defining your Evaluation Period
 Time period: You need to decide how much data you need to
analyze when computing the average or peak values to make
sure that it captures your busy periods or is a good
representation of your business cycles.
 Priority: Many administrators will divide up their systems into
high, medium and low priorities with different metrics for each
group.
Computing Resources Consumed
The next step is to look at each resource and compute the average
or peak utilization for your evaluation period.
Generate Recommendations
The next step is to put all the pieces together and evaluate each
resource for each VM and determine if the right amount of CPU,
Memory and Storage has been assigned to the VM.
Wasted Storage
• Easy to create VMs
• VMs are large files
• Not always easy to find any files that are not being
used.
Types of Wasted Storage
• Abandoned VMs – A virtual machine file that is on
your datastore but is not attached to a VM listed in
vCenter or the host.
• Powered off VMs
• Templates
Not used in 30 days or more
• Snapshots
• Zombie VMs – A virtual machine that is running but
not being used.
Wasted Storage
Abandoned
VMs
Powered
off VMs
Manual or Automated?
VKernel’s Optimization Pack and Capacity Analyzer does this work
for you.
Conclusion
The goal of virtualization is to find the balance between
correctly sizing your environment while achieving
maximum performance with the least amount of
resources.
Download a trial of VKernel’s appliances to resolve
Performance bottlenecks and optimize your
environment!
http://www.vkernel.com