This is a presentation ServiceMesh's Dave Roberts did for a Focus webinar, March 2011:
http://www.focus.com/webcasts/information-technology/private-cloud-right-your-organization/
1. THE AGILE IT PLATFORM
Is Private Cloud Right for Your Organization?
A 5-Step Litmus Test
Dave Roberts
Vice President, Strategy
dave.roberts@servicemesh.com
Company Twitter: @servicemesh
Personal Twitter: @sandhillstrat
2. Who is ServiceMesh?
Provider of Agile IT governance, security and lifecycle management software and services
for Global 2000 clients
Customers include some of the world’s largest and most sophisticated companies in:
Financial services
Health care
Consumer
Other IT-intensive industries
Global presence with headquarters in Los Angeles and offices in Austin, London,
New York, Sydney, and Washington D.C.
3X revenue growth in 2010
3rd consecutive year of profitability
2
3. “I Heard There’s No Such Thing as Private Cloud”
• Some large cloud service providers want to force the world into a one-size-fits-
all, public cloud model.
• Anything that challenges that public cloud world-view must be attacked.
• FACT: Every G2000 enterprise that ServiceMesh is working with is building
private clouds.
• FACT: The TM Forum’s Enterprise Cloud Leadership Council (ECLC) has defined
requirements and a working group that encompasses private cloud (enterprise-
grade IaaS).
3
4. The World is Multi-Dimensional
Excess
Public Capacity AWS EC2
Sellback
Internal IT
Private or Savvis
Service Provider
Internal External
4
5. Private vs. Virtual Private
• Public cloud vendors are retrenching and offering “virtual private cloud”
• Characteristics
– Shared infrastructure
– Private access via VPN
– Rich networking functionality
• Private IP address space makes it appear as part of the enterprise data center
• Question: Why would they do such a thing if private cloud didn’t exist?
• Answer: Because enterprises are demanding more than public clouds.
But they want to leverage the infrastructure across multiple customers for
higher efficiency.
5
6. Hybrid Clouds Enable Optimization
Optimize workload placement and schedule
Enable portability across clouds
Align best fit providers to unique needs
Drive price arbitrage between providers
Virtual Dev and test Grid nodes
desktops Grid nodes Dev and test
environments (Night)
(Day) (Day) environments
Internal Private Cloud External Public Cloud
6
7. Private Cloud Litmus Test
Do you have specific security
1 requirements?
• Data in transit
• Data at rest
• Data destruction requirements
• Audit requirements
• Regulated industry
7
8. Private Cloud Litmus Test
Are you of sufficient scale?
2
• Doesn’t make sense to build an optimized
private cloud if you’re small
• But don’t underestimate yourself
• If you have 500 workloads or more, It
might make sense to build a private cloud
8
9. Private Cloud Litmus Test
Do you have a reasonable fraction of
3 dynamic workloads?
• If your workloads are largely static, then
you might be better off with traditional
server or virtualization solutions
• You might still want to outsource in a
traditional managed hosting model
• Or go hybrid here:
“Buy the base, rent the peak”
9
10. Private Cloud Litmus Test
Can you deal with the organizational
4 dynamics associated with the effort?
• Building a private cloud is not simply a
matter of purchasing and assembling
technology
• Shifting to a cloud operating model will
require changes to processes and job roles
• You’ll need sufficient top-down support to
effect lasting change
10
11. Private Cloud Litmus Test
5 Do you have special requirements?
• Predictable outcomes
• Reserved capacity
• Known performance (CPU, disk, network)
• Known latencies
11
12. Basic Private Cloud Architecture Stack
Private Cloud Users
Billing
Workflow
Policy and Governance
Automation
Security
Cloud
API
Hypervisor Private External
Cloud Public
CPU Storage Network Pod Cloud
Private Cloud Pod
12
13. Pod Reference Implementation
Infrastructure selection will vary based
on specific organizational needs to
provide the optimal combination of
performance, cost, and simplicity.
Look for a balanced design that delivers VM density:
the right ratio of CPU, storage, and 1600 server workloads
per compute rack
network bandwidth. 6000 desktop workloads
per compute rack
48 blades for a total of
576 cores. Standard
memory config is 96
GB/blade, high end Two 10 Gbps
config is 192 GB/blade. switches, each
with 480 Gbps
of network
bandwidth.
100 TB of raw
storage capacity
and 500 GB of
memory. Other reference implementation metrics
and technical specs are available.
13
14. The Need for Balance
CPU
Networking
Storage
Memory
• Want 1 – 2 GHz • 100 GB – 500 • Server: 3 – 4 GB • 2 – 4 10GE ports
per VM GB per VM • VDI: 1 – 2 GB per blade
• Servers: 2 VMs • Server: 100 IOPS
per core • VID: 30 IOPS
• VDI: 10 VMs per
core
14
15. Pod Reference Implementation
Based on published external cloud
prices, financial analysis showed that
the private cloud can cost up to 75%
less than comparable external cloud
offerings for a high-availability use-
case.
End-to-end automation required
only minutes to provision workload,
auto-scale, and configure security
for complex topologies, which
normally would take days or weeks.
15
16. Adopt a Cloud Operating Model
• Private cloud is not a product or technology you can simply buy and install
– FAQ: “What’s the difference between virtualization and private cloud?”
– Answer: Virtualization is a technology used to build private clouds, but clouds are
more about how you operate the technology and expose it to your users than the
technologies themselves
• To extract the full advantage, you’ll need to run your internal private cloud like a
(mini) service provider
– That means you’ll have to move from your traditional IT operating model to a cloud
operating model
• To implement a cloud operating model, you’ll need to:
– Rework processes and procedures
– Define internal SKUs for your users
– Create SLAs
– Create internal pricing
16
17. Creating Internal Private Cloud SKUs
Private Cloud SKU Price Don’t forget this!
Use it to fund
Internal Markup maintenance, upgra
des, continual
development, and
SKU Cost support.
SKU Capacity Think like a service
provider. Run it like a
Aggregate Pod Capacity business!
17
18. Understanding Contestability
• Explicit Goal: Make your private cloud earn its keep.
• To do this, you need to seek opportunities for “contestability”
• You want competition between your own private offering and others
• Other offerings might be other internal private clouds!
• Don’t create a private cloud and run it poorly!
• Don’t waste money on something that isn’t competitive
• Don’t settle for poor service
18
19. Private Cloud Results
Deploy Fast: Scale Faster:
Provisioning with private cloud: 4 minutes Provisioning Time*
25
20
15
Minutes
Self-service Automated policy Start using it 10
portal enforcement
5
Provisioning with virtualization only: 3-6 weeks 0
1 25 50
Number of simultaneous server
VM IT Provision VM Start using it instances provisioned
request approval and configure
Provisioning with hardware procurement: 12-18 weeks
Hardware Procurement Hardware Rack in data Setup platform/ Start using it
requisition approval arrives center application
* Reference implementation test results, Nov 2010 19
Editor's Notes
Elevator Pitch: ServiceMesh has been around since 2008 (really late 2007). If you wanted to really to simplify it… we’re a software company in the Cloud Mgmt space… although that’s a very limiting description. More accurately, what ServiceMesh really does is help large enterprises adopt what we call “Agile IT operating models”. (Cloud is part of this…but its much more than just cloud… which I’ll explain shortly). What this allows them to do is drastically reduce the money and resources consumed by their current IT operations and infrastructure, and instead redirect it so IT can become more innovative, more adaptable to business change, and even enable new business models. The way that we do that is by providing advisory services and a product platform that fully governs and manages federations of SaaS, PaaS, internal and external clouds providers to create an “everything-as-a-service” IT environment that is more agile and cost effective than what enterprises are using today. Our GTM approach is to lead with a series of (what we call) pre-packaged IT optimization strategies, that are basically solution accelerators, that provide a low-risk entry point…. and customer build upon these as stepping stones in a path toward this larger operating model transformation.As a company, we’ve been working on these challenges since 2008 with large enterprise customers…before cloud computing became such an overused buzzword. And we feel fortunate to have gotten a good head start on where IT is headed today with cloud… particularly from an enterprise perspective. Our customers are Global 2000… most with very complex, highly regulated, competitive environments….. Some with annual IT budgets exceeding $1B.The ServiceMesh team itself has acquired expertise in several areas working with these enterprise accounts. Not just in cloud technologies… but also IT strategy, architecture and design, platform integration, security, governance, organizational change management, and so on.ServiceMesh itself is a global company. We have customers in NA, EMEA, and AP. Austin is Dev center. LA contains a good portion of the exec mgmt team.
The considerable governance and lifecycle mgmt capabilities in the Agility Platform enable many of the core capabilities we just discussed. Those same capabilities are also critical for other cloud use cases. The Agility Platform is designed to provide a unified platform to govern and manage all of them. This provides additional benefits to manage the placement and schedule of all of these workload deployments.
Cost per VMCost comparisons to EC2Capacity unit optimizationVM density optimizationAutomated provisioningAuto-scalingDynamic security configurationThe physical infrastructure that include CPU, network, storage, and memory. The virtual environment runs on top of the CHP hardware and supports the IaaS and IPaaS offerings with interface with the security and management layer The CHP defines an optimised, standardised, high density, virtual infrastructure hosting configuration which is designed to meet price /performance benchmarks. CHP is optimized forPricePerformance (both bare metal and virtualized hosting)Rapid deploymentReduced complexityOptimizedfor virtualization, stateless computing, and hosting a large number of instancesDesigned for high density, low power and minimal heatThe same model can be used both internally and at external hosting providers, but it is not a requirement for the external hosting provider to utilize the architecture in order to provide computing supportThe IaaS offering can be either internal hosted or external hosted, and the compute support offering is the same under either model. Under the external private compute model, the hosting provider owns, manages and controls the underlying infrastructure, but CBA retains control over the OS, storage and deployed applications.CHP is deployed in modular units on standardised commodity hardware.The CHP defines an optimised, standardised, high density, virtual infrastructure hosting configuration which is designed to meet price /performance benchmarks. It does not necessarily define specific virtualisation software or underlying hardware manufactures.
Cost per VMCost comparisons to EC2Capacity unit optimizationVM density optimizationAutomated provisioningAuto-scalingDynamic security configurationThe physical infrastructure that include CPU, network, storage, and memory. The virtual environment runs on top of the CHP hardware and supports the IaaS and IPaaS offerings with interface with the security and management layer The CHP defines an optimised, standardised, high density, virtual infrastructure hosting configuration which is designed to meet price /performance benchmarks. CHP is optimized forPricePerformance (both bare metal and virtualized hosting)Rapid deploymentReduced complexityOptimizedfor virtualization, stateless computing, and hosting a large number of instancesDesigned for high density, low power and minimal heatThe same model can be used both internally and at external hosting providers, but it is not a requirement for the external hosting provider to utilize the architecture in order to provide computing supportThe IaaS offering can be either internal hosted or external hosted, and the compute support offering is the same under either model. Under the external private compute model, the hosting provider owns, manages and controls the underlying infrastructure, but CBA retains control over the OS, storage and deployed applications.CHP is deployed in modular units on standardised commodity hardware.The CHP defines an optimised, standardised, high density, virtual infrastructure hosting configuration which is designed to meet price /performance benchmarks. It does not necessarily define specific virtualisation software or underlying hardware manufactures.
The physical infrastructure that include CPU, network, storage, and memory. The virtual environment runs on top of the CHP hardware and supports the IaaS and IPaaS offerings with interface with the security and management layer The CHP defines an optimised, standardised, high density, virtual infrastructure hosting configuration which is designed to meet price /performance benchmarks. CHP is optimized forPricePerformance (both bare metal and virtualized hosting)Rapid deploymentReduced complexityOptimized for virtualization, stateless computing, and hosting a large number of instancesDesigned for high density, low power and minimal heatThe same model can be used both internally and at external hosting providers, but it is not a requirement for the external hosting provider to utilize the architecture in order to provide computing supportThe IaaS offering can be either internal hosted or external hosted, and the compute support offering is the same under either model. Under the external private compute model, the hosting provider owns, manages and controls the underlying infrastructure, but CBA retains control over the OS, storage and deployed applications.CHP is deployed in modular units on standardised commodity hardware.The CHP defines an optimised, standardised, high density, virtual infrastructure hosting configuration which is designed to meet price /performance benchmarks. It does not necessarily define specific virtualisation software or underlying hardware manufactures.