Cloud Computing - A Practical View

Mandeep Dhami
http://geekandpoke.typepad.com/geekandpoke/2009/03/let-the-clouds-make-your-life-easier.html
Overview
• The Context
– A specific project scenario

• Why Cloud Computing?

– Economic drivers
– Flexibility and agility
– New capabilities

• Why not Cloud Computing?

– Regulatory constraints
– Operational concerns
– Technical issues

• And the Practical “Middle Way”!

– Services evaluated
– Proposed engagement
The Context
• Cloud computing can mean • In this talk we evaluate the
different things to different trade-offs in context of the
people following hypothetical scenario:
– You work on a medicare/medicaid
eligibility system
– Field workers use a web based
tool to input case details and to
check status
– Web server is implemented using
java/websphere on a Windows
Server
– Backend eligibility sub-system is
implemented using COBOL on a
IBM mainframe
– You are tasked with evaluating a
cloud based solution for the web
tool
http://www.nature.com/ki/journal/v62/n5/fig_tab/4493262f1.html
Many Layers of the Cloud
Some Initial Design Constraints
• Type of cloud service required - IaaS or Private Cloud
– Since it is a custom software application, SaaS is not an option
– Since the platform is also very custom (for libraries and versions) and has some
non-standard libraries (say websphere v6.5, DB2 v9.1, JCA for CICS, etc …),
PaaS is not an option either.
– IaaS might be feasible as we own the software stack in that model
– Private cloud can always be used, as we will own the cloud in that model!

• Type of connectivity required – “VPN to VM”

– We will need secure encrypted connection to backend system for the web
application to get/update case status. Conceptually this is like a VPN from the
VM to the backend.
– Any IaaS solution that does not provide secure connection from the server VM to
internal LAN can not be used
Why Cloud Computing?

To cloud or not to cloud, that is the question …

http://geekandpoke.typepad.com/geekandpoke/2009/11/simply-explained-project-risk-update.html
Economic Drivers
• Pay as you go • Lower support costs
– No upfront cost to acquire – The team does not have manage
server/network hardware hardware, network or storage for
– Only pay for dev and test systems production system
during dev and test phases – No need to hire expensive
– No upfront cost to “try” new consultants for non-core
features like Web Firewalls (infrastructure related) activities

• Deterministic Project Costing • Lower hardware costs

– More transparency regarding – Typical server utilization is low,
infrastructure costs pay only for what you use
– Less risk from last minute capital – Typical network utilization is low
cost request related to production (routers, firewall, etc), pay only
usage for what you use
– Not encumbered by internal
transfer accounting!
Flexibility and Agility
• Rapid Scaling • Dynamic Provisioning
– Start small, scale as required – Spin up more test-beds as
based on production performance required. Keep test execution
measurements moving even as developers are
– Respond faster to customer debugging on an existing setup
demand for capacity – Spin up systems to do load testing
– Respond faster to features that as required. Pay only for the time
require more compute/storage used to do the tests
resources

• Dynamic Infrastructure • More Choice

– Enable infrastructure changes with – Change infrastructure vendors for
mouse clicks better SLA or price without
– Increase server pool for batch impacting/altering the application
processing as required – meet any – Do “Beta test” for a few case
batch window (at some cost) workers on a small system, roll
– Developers can prototype “at out new code incrementally
production scale” and capacity – Roll back to a previous image, as
a fallback option
New Capabilities
• Next Gen architectures • Accelerate innovation
– Enable disaster recovery by using – Shift from supporting the
a service provider with multiple infrastructure to innovating on
physical locations application
– “Try” new features like – Use cost transparency to innovate
memcached, CDNs, etc. without processes and reduce waste
new investment in hardware or
infrastructure expertise

• Advanced infrastructure • Green computing

capabilities – Increase server utilization, reduce
– Change management to server power usage
configuration is centrally managed – Use more efficient cooling, reduce
and encapsulated power usage
– Self healing, hot backups etc. – Reduce number of servers and
available reduce waste
– API’s available to infrastructure
for flow-thru’ automation
Why Not Cloud Computing?

There be dragons …
First, you sometimes hear some FUD …

“We will have no liability to you for any unauthorized access or use,
corruption, deletion, destruction or loss of Your Content or
Applications”
Customer Agreement, Amazon Web Services

“Salesforce.com shall not be responsible or liable for the deletion,

correction, destruction, damage, loss or failure to store any
customer data”
Master Subscription Agreement, Salesforce.com

… but this is not really very different from software EULA

(So we believe that you can safely ignore this issue, except during
during contract negotiation)
But there are Real Regulatory Constraints
• Privacy • Forensics and audit
– Since this project handles medical – If your cloud APIs can not be
data, HIPPA rules apply audited for forensic investigation,
– If your cloud infrastructure can you can not use it for sensitive
not be HIPPA compliant, you can data
not use it – If audit data is not
cryptographically secure, it lacks
adequate controls

• Governance mandate • PKI infrastructure

– Just because the application is on – How are private keys stored and
cloud, the governance mandates managed by the cloud based VMs?
do not go away! – Can you meet FIPS requirements
– Can you produce reports on usage that you currently meet with
or controls that are comparable to hardware/physical security
a system with physical security? constraints?
And Real Operational Concerns
• The Blame game • Priority management
– When there is a problem today, it – When you have a customer
is already painful to get from situation, your “tech team” works
defect to defect ownership … on it as #1 priority till it is
resolved …
When a problems occur in cloud,
how do you get from the “conf-call How do you set priority for the
from hell” discussing defect to cloud vendor’s tech team to fix
productive “root cause analysis” your specific problem among their
and taking defect ownership? priorities?

• SLA “assurance” • Vendor lock-in

– Can you measure service levels in – How real is the promise of choice?
terms of the metrics used in the – To resolve the technical or
SLA in the contract? operational issues, are you tying
– Do you get reports on “real SLA” into a proprietary API that limits
or on a synthetic benchmark? any real choice?
– Do you get “continuous reporting”
of metrics that you can use for
trend analysis and planning?
And Very Real Technical Issues
• Visibility • Security
– Clear system boundary with – Encrypted VPN from “Server VM to
adequate instrumentation the Backend network”
– Tools to view infrastructure usage – SSO integration for admin/API
by your application usage
– “Safe sharing” of shared resources
(like network, swap, crash dump,
etc).

• Diagnostics • Network Services

– On demand capture of data, traffic – No good model for application
and performance statistics level network services (like
– Flow thru’ integration with firewall, load balancer, etc)
automation/tools – We can use x86 VMs as virtual
– Automated data capture (black appliances, but they lack the
box) before the VM image is lost. hardware acceleration of typical
network devices
The Practical “Middle Way”

In Buddhism, the “Middle Way” is the Nirvana-bound path of

moderation - away from the extremes of sensual indulgence and
self-mortification and toward the practice of wisdom, morality and
mental cultivation.
From http://en.wikipedia.org/wiki/Middle_way
http://en.wikipedia.org/wiki/Middle_way
 From http://dilbert.com/strips/comic/2009-
http://dilbert.com/strips/comic/2009-11-
11-18

… No I really did not mean that!

Cloud Service’s Evaluation for This Specific Project
NOTE: This is a sample evaluation. Your results will differ based on the assumptions
that you make on the project and on the services them selves

Service Product Regulatory Operational Technical

Provider Constraints Concerns* Issues

Amazon EC2

Solid performer, lots of 3rd party support

Rackspace Mosso

Solid performer, good enterprise support

Savvis Virtualization in the Cloud

Closest to a private cloud (VMware), very

good enterprise support

Appnexus Appnexus Cloud

Not clear how it will handle issues specific to

government or HIPPA compliance

* Assuming appropriate relationship and contract/penalties

Engagement Proposed for This Specific Project

• First qualify the service provider’s offering for regulatory issues

– HIPPA
– PCI (if you accept credit cards for fees)
– FIPS (for PKI)
– Etc

• Then qualify your relationship with the service provider so that you can handle
operational issues around “blame game”, priority management etc.

• Then qualify the network, the virtual servers, and the storage for security, visibility,
manageability, diagnostics, etc. In particular, qualify the secure VPN to your virtual
servers (like Amazon’s VDC)

• Finally move development and test of next major upgrade to cloud service provider.
Do a beta roll out first, and then scale incrementally as you build confidence.

• With dev & test success behind you, use it as a model to transition the production
servers (for the web application) to the cloud.

• Always, incremental build-up based on success of the previous step!