Sl virtual apps-131106

© 2012 SL Corporation. All Rights Reserved.

© 2013 SL Corporation. All Rights Reserved.1

Shortfalls in Monitoring Virtualized Apps Deficiencies and Opportunities

Tom Lubinski, CEO, SL Corporation

6/7 November 2013



Shortfalls in Monitoring Virtualized Apps

• What do we mean by “Virtualized Apps” ?

• What are the “shortfalls”

• Deficiencies

• Opportunities



Shortfalls in Monitoring Virtualized Apps

Another way to look at our agenda:

• Hope springs eternal in the Human breast– the promise

• Be careful you don’t go from the frying pan into the fire– the reality

• Everybody wants to save the earth; nobody wants to help Mom do the dishes – P.J. O’Rourke

– what needs to be done



Virtualization: The Promise

Ready and Easy Access to Infrastructure …

Success !




Automatic Scaling of Applications …

Not so successful … yet




Virtualization Changes Everything …

Hmmm … not so fast



Virtualization: The Reality

See how far we have come …

A Virtual Machine doesn’t look that much different from a physical machine at this level …

1990

Terminal + BSD Unix + vi

2013

putty + Linux + vi + colored fonts !!




We’ve gone from …

Timesharing on an IBM Mainframe Computer




To …

Timesharing on an ESXi Hypervisor

… on a smaller box, of course




Problem #1 – Hypervisor Overhead

There is a cost associated with slicing CPU time and resources for each VM

Sometimes called “Cpu Steal” … can be > 10%

See Patrick Eaton’s blog: http://www.stackdriver.com/understanding-cpu-steal-experiment/




Problem #2 – Noisy Neighbors

You don’t have the CPU to yourself !

If a “neighbor” is using a large part of the Host CPU, your performance will be affected

Result = poor response time

See Bernd Harzog’s blog: http://www.virtualizationpractice.com/applications-performance-equals-response-time-not-resource-utilization-9916




Problem #3 – Resource Exhaustion

You don’t have the Network and Storage to yourself !

Network-dependent apps can have highly unpredictable response time

Storage-centric apps can be impacted by limited read/write capacity




Problem #4 – Poor Capacity Utilization (my favorite …)

Expectation: Virtualization = better resource utilization

Problem: Too easy to say - “let’s just fire up another VM”

You can “see and touch” hardware

Virtual Machines get lost in the aether




Deficiencies in Monitoring …

Cannot just monitor the CPU of the VM itself; must know what the Host and hypervisor are doing

No direct correlation between VM resource utilization and application performance

System Management tools are NOT part of platform provisioning; monitoring must accompany the application

Distributed nature of VM deployment means that traditional agent-based tools don’t work



Virtualization: The Opportunity

Good News:

The shift to virtualization provides an opportunity to improve on the old ways of monitoring applications

Let’s look more closely at what is meant by “virtualized applications”



Virtualization: The Opportunity (1)

Level One … Survival !

Adapt Monitoring to Virtualized Environments:

Use agentless data collection and/or agents that communicate data across firewalls or over public internet

Include data from Hypervisor in health state and performance calculations (cpu / mem / disk / network)

Include application-related metrics such as response time in HS and Perf calculations




Level Two … Accomplishment !

Provide Complete Visibility into Health State of Virtualized Apps:

Collect ALL data relevant to every application

(infrastructure, middleware, application metrics)

Normalize data so it can be aggregated and correlated across all data sources and types

Provide application-centric “Views” of monitoring data relevant to each application




Collect ALL data relevant to every application

Each application is typically dependent on multiple component types

This includes both VMWARE-VM AND VMWARE-HOST

Health State of App is the aggregate of all components weighted by criticality




Provide application-centric “Views”

Aggregate health state of each application propagates up to higher-level summary views

Single alert generated for an application independent of number of alerts in components




Including History can provide longer-term views of app behavior




Level Three … Mastery and Automation !

Use Provisioning Data to Automate Monitoring

Configure all Monitoring at time of Provisioning

(define target connections, operating parameters)

Configure component dependencies based on component type

(WebLogic App runs on WebLogic server, e.g.)

Populate Service Dependency Model from Provisioning Metadata

(dynamic CMDB)




Configure all Monitoring Connections at time of Provisioning …

Infrastructure as a Service – IaaS (well-defined)

Provision VMs, Storage, Network Resources

metadata readily available for use in CMDB (vSphere, e.g.)

and make connections to collect monitoring data

Platform as a Service – PaaS (a little more vague)

Provision Middleware on which Apps are built metadata needs to be extracted for use in CMDB

when deploying JMS servers or App Servers (e.g. IBM SmartCloud)

some custom development in this area




Determining component dependencies based on component type

VMWARE-HOST

VMWARE-VM VMWARE-VM

WLS WLS

WLS-CLUSTERAPP

WLS-APP WLS-APP

OC-CACHEOC-CLUSTER

OC-CACHE OC-CACHE

Here, all components are determined from the single entity: WLS-CLUSTERAPP …




Examples of “defined dependencies” …

WLS-CLUSTERAPP WLS-APP

WLS-CLUSTERWLS

OC-CACHE

VMWARE-VMVMWARE-HOST

OC-CLUSTER

TOMCAT-APP TOMCATHOSTJVM

JMS-TOPIC JMS-SERVERVMWARE-VM

VMWARE-HOST




Generate table of dependencies in CMDB from single entry !




An Even Better Approach: (Future !)

Generate all dependencies during Application Provisioning

User Demand

Provisioning System

Config Metadata

Deployed Application + Infrastructure

Monitoring System

In a sense, the provisioning metadata is the CMDB of the future …



Monitoring Virtualized Apps

Some Conclusions …

Virtualization DOES change things … adds complexity

Deficiencies in traditional monitoring approaches require new tools and processes – survival first

Opportunity exists when virtualizing apps to upgrade monitoring approach

Powerful techniques can be used to automate much of what used to be very manual



How RTView Addresses This Problem

www.sl.com www.sl.com/blog

http://www.sl.com/

http://www.sl.com/

http://www.sl.com/blog



Thank you!For more information, please visit

www.sl.com www.sl.com/blog

http://www.sl.com/

http://www.sl.com/

http://www.sl.com/blog



Sl virtual apps-131106

Technology

monitoring virtualized apps

virtualized apps

health state

data relevant

vm vmware

component type

rights reserved

app oc