1. Infrastructure-as-a-Service (IaaS)
Cloud Computing, CS
Infrastructure-as-a-Service (IaaS)

2. Outline Cloud Architecture Infrastructure-as-a-Service (IaaS)
Motivation
Overview
The IaaS Stack
Migration Strategy
What to Watch for?
VM for IaaS Environment
Infrastructure Services
Effect of Virtualization & Cloud Computing on Data Centers
Summary and Conclusion
I can pretty much read this one straight through and provide details in following slides.

3. Cloud Architecture
Here’s an overview of cloud computing. Basically the main bulk of computing power resides in the cloud, accessible on-demand through network. There may be a separate software/service provider or a cloud user that channels/aggregates/provisions the desired services to end users.

4. Cloud Architecture e.g., Web browser SaaS , e.g., Google Docs
PaaS, e.g., Google AppEngine
IaaS, e.g., Amazon EC2
Here’s an overview of cloud computing. Basically the main bulk of computing power resides in the cloud, accessible on-demand through network. There may be a separate software/service provider or a cloud user that channels/aggregates/provisions the desired services to end users.

5. Cloud Architecture Amazon EC2
Clients can rent virtualized hardware, can control the software stack on the rented machines
Google AppEngine
Provides a programmable platform that can scale easily
Microsoft Azure
Clients can choose languages, but can’t change the operating system or runtime
IaaS
PaaS
Here’s an overview of cloud computing. Basically the main bulk of computing power resides in the cloud, accessible on-demand through network. There may be a separate software/service provider or a cloud user that channels/aggregates/provisions the desired services to end users.

6. IaaS: Motivation - On-Premise Application Lifecycle
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

7. IaaS: Overview
IaaS Platform should allow you to automatically provision servers, storage, and network elements with flexible billing options
IaaS should provide “building blocks” that allow you to shape your infrastructure to fit your changing needs of your business and to respond quickly as needs change
Benefits:
No upfront cost
Multi-tier auto provisioning within hours
Scalability, services/resources are provided on demand
Flexible/Secure direct connectivity
SLA with 99.9% uptime
In general, cloud computing can mean 1. applications/services delivered over the internet 2. HW and Systems SW in DC that enable the services.
Features: Centralization- think of Office, OS, AV updates. Roughly, there are 3 service models.

8. IaaS: Overview
IaaS provides an environment for running user built virtualized systems in the cloud
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

9. IaaS: The Stack Many players; the cloud is just beginning
In the future, expect to see all large vendors riding on the complete stack
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

10. IaaS: Migration Strategy: Staged or Direct Migration
Virtualized Instance
Hardware
Public Cloud
Private Cloud
V2V
P2V
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

11. IaaS: What to Watch For? We all have lots to learn
Check your security policies can be satisfied by the Cloud provider
Does the SLA meet availability requirements?
You are now porting your OS and upper stack
You will need to maintain it
Remember the Cloud is its infancy
It’s immature
We all have lots to learn
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

12. IaaS: Building VM for an IaaS Environment
Runtime environment is shown to the RHS
Once a VM is uploaded to an IaaS Environment, it can be configured to use the IaaS’s raw storage
VM is ready to run and the application within it
Building application to run in an IaaS is similar to building it to run on premise
Developers use familiar tools, programming languages, and any needed licensed products (DBMS, workflow, etc.)
Once app development is completed, the application and any licensed software are installed on a VM; VM is thought of as a container that insulate the app from the runtime environment
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

13. IaaS: Inside VM Built for an IaaS Environment
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

14. IaaS: Infrastructure Services
IaaS can be viewed as three categories: Servers, Storage and Connectivity
Cloud Providers would offer virtual Server instances on which the customer can install and run a custom image. Persistent storage is a separate service which the customer can purchase. Finally, several offerings for extending connectivity options
Amazon is the de-facto standard for IaaS. Most IaaS offerings are either complements to Amazon Web Services (AWS) or competitive to them
There is an open source equivalent to AWS and roughly compatible to its APIs, Eucalyptus (Elastic Utility Computing Architecture for Linking Your Programs To Useful Systems); shipped with Ubuntu since version 9.4
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

15. IaaS: Infrastructure Services
Servers:
Servers represent compute resources along with minimal storage and IO channels
Server outsourcing model is divided into three allocation options: (1) Physical, (2) Dedicated Virtual, and (3) Shared Virtual
Physical means dedicated hardware. Dedicated Virtual means dedicated hardware but with a hypervisor so the customer can run multiple OS. Shared Virtual are exposed to customers as pool of VMs
Amazon is Shared Virtual, each is based on Amazon Machine Image (AMI). Customer can use pre-packaged AMIs from Amazon or build their own
A challenge for Virtual servers is that they do not maintain local storage; any configuration information has to be stored externally
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

16. IaaS: Infrastructure Services
Some offerings such as Rackspace Cloud offers local persistent storage as well as large pool of dedicated IP addresses on all instances
Storage:
Moving from storage Networks (SAN) to Storage-as-a-Service. It is characterized by a location-agnostic, virtualized data store with the illusion of infinite capacity while easy to use.
Storage services are also useful for archiving, content delivery, disaster recovery, and web application development
To cater to cloud, vendors like EMC with their Atoms product, have begun to deliver h/w and s/w that is specifically designed for geographically dispersed content depots with replication, versioning, de-duplication, and compression capabilities
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

17. IaaS: Infrastructure Services
On-demand storage services, relevant factors include persistence, replication, and access speed and latency
Due to synchronization requirements of Content Delivery Networks (CDN - Akamai), you may observe very different speeds for Read and Write data
Amazon offers two persistent storage capabilities: (1) S3 (Simple Storage Service) and (2) EBS (Elastic Block Storage)
Amazon AMI do not have persistent storage but locally mounted disks can be used for logs, interim data, etc.
S3 is accessible through REST and SOAP APIs. It offers distributed, redundant buckets that are replicated using Amazon’s CloudFront CDN across Europe, Asia, and US.
S3 can accommodate data sizes from 1 Byte – 5GB and provide permissions for access control based on AWS authentication
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

18. IaaS: Infrastructure Services
S3 provides versioning so that customers can recover accidently deleted or overwritten objects
EBS is intended for high-performance virtual hard disk. It can be formatted as a file system and then mounted on any EC2 instance. The size can range from 1 GB – 1 TB.
Amazon provides a mechanism to store an EBS snapshot in S3 for long term durability
Other storage services include:
The Rackspace Cloud: similar to S3 with REST Access API
GoGrid: no access API; mechanisms such as secure Copy (SCP), FTP, Samba and rsync
LayeredTech: The Dynavol service supports mirroring and many access mechanisms in cluding:HTTP, FTP, SFTP, SSH, SCP, rdiff, rdist, rsync, and smbfs
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

19. IaaS: Infrastructure Services
Hadoop HDFS: partitions large files across several network hosts. It delivers blocks of data over the network using propitiatory protocol or HTTP for browser access. Data nodes communicate with each other to rebalance data and to maintain replication of data high (default is 3)
Amazon also provides Cloud Database (Amazon Simple DB) for structured data. While data is tabular, it is not conformant and doe not support JOIN, foreign keys, triggers, and stored procedures
Simple DB does not use schema; instead it defines “domains” with items that consist of up to 256 attributes and values. Value can be from 1B to 1KB. Simple DB supports simple operators. Queries are confined to a single domain
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

20. IaaS: Infrastructure Services
Simple DB does not use schema; instead it defines “domains” with items that consist of up to 256 attributes and values. Value can be from 1B to 1KB. Simple DB supports simple operators. Queries are confined to a single domain
Other interesting data services include:
Google Big Table: extremely large scale DBMS in the PB range
Dynamo: Amazon’s version of Big Table
Cassandra: Facebook distributed storage system
Network:
Amazon EC2 instance will receive single DHCP (Dynamic Host Configuration Protocol) address. Additional addresses, static or persistent domain addresses, can be requested separately
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

21. IaaS: Infrastructure Services
Two other network-related functions are available: (1) provisioning for network segmentation and mechanism to bridge segments. (2) Load balancing.
Amazon allow customers to define firewalls which restrict the inbound/outbound traffic to specific IP range and port numbers
Other Cloud providers, such as AppNexus, provides each customer with a private VLAN, i.e., avoids being exposed to broadcast traffic
Amazon Virtual Private Cloud, 2009, enhances and secures connectivity between cloud services. It allows an Enterprise to create a VPN between the Enterprise and Amazon AWS
Once connectivity is established, the next task is ensuring that it performs! Two factors defining network performance: (1) latency. (2) Throughput or bandwidth available from the network
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

22. IaaS: Infrastructure Services
Many providers offer Load Balancing capability. One option is to use a DNS Round-Robin that involves adding multiple IP addresses to the same DNS Domain. Alternatively is to use either software or hardware load balancer
Integration:
After setting the network connectivity, next step is to configure the application to be able to exchange data. In theory, IaaS is not the right level to facilitate integration; it is rather an added value provided by many of the Cloud providers
Amazon provides Simple Queue Service (SQS): it is unlimited number of queues and messages with message size is up to 8 KB.
Apache Hadoop provides a framework for tightly coordinated interaction of applications (Job Tracker, Task Tracker)
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

23. IaaS: Infrastructure Services
Management:
IaaS components are managed through the vendor virtualization facilities. Depending on the hypervisor used, there are 3rd party and native components for Xen, Vmware, Hyper-V, KVM, etc.
In addition, there may be an umbrella management framework such as BMC Patrol, CA Unicenter, IBM’s Tivoli, etc.
For Public Cloud, there are services that help manage established infrastructure providers: Elastra, RightScale, enStratus, etc. which provides F/Es for managing Amazon EC2 as well as other IaaS offerings
Payment and Billing:
Monetization of utilities-based services relies on: (1) fine grained metering. (2) Simple and trustworthy payment system
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

24. IaaS: Infrastructure Services
Many Cloud offerings are from international communication giants such as BT, NTT, etc.
eCommerce companies are also visible in that space such as PayPal
Still many Cloud providers developed their own instrumentation and billing system
IaaS is frequently used as SaaS enabler, i.e., not providing only raw compute, storage, and network resources but also help to facilitate SaaS with payment and billing services
Amazon Flexible Payments Service (FPS) is a service for developers, that leverages Amazon’s sophisticated retail billing system. The customer can use the same identity, shipping details and payment information as they would for ordering directly with Amazon
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

25. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Multi-tier hierarchical networking structure: Typically 3-4 layers including access layer switches, aggregation switches, and core switches. These core switches usually contain the vast majority of the intelligence in the network
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

26. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Today’s applications are more distributed, often multi-tiers, and are oriented toward service delivery
The above applications architecture results in:
Greater network traffic on the Ethernet network, including storage traffic such as iSCCI
More storage traffic as applications use distributed file system
Higher server-2-server traffic or VM-2-VM traffic
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

27. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Limitations of hierarchical networking structure: The more network traffic, the more limitations with hierarchical networks including:
Spanning Tree Protocol (STP) Limitations: STP detects and prevents loops in L2 networks by allowing only one active path from one switch to another
Oversubscription: is an issue in the aggregation layer, e.g. Cisco Universal Computing System (UCS) may have oversubscription rates of anywhere from 4:1 to 32:1 in the aggregation layer!
Port Extension Technology: IEEE is developing a “port extender,” which is physical switch with limited functionality
Latency: can be more of a challenge than oversubscription and row bandwidth; due to the nature of traffic within an application
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

28. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

29. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Changing Business Applications:
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

30. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Server Virtualization (VM):
Increase in VM density and with more powerful CPU & memory subsystems resulted in greater performance demand on the network subsystem at the server-network edge.
Moving workloads dynamically requires VMs to stay within a common VLAN in the same layer-2
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

31. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

32. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Client Virtualization (Virtualized Desktop Infrastructure - VDI):
a specialized type of VM is the client virtualization technology such as Virtual Desktop Infrastructure (VDI). VDI creates a client desktop as a VM. The VDI instance is more than simple VM, it includes real-time compilation of the end-user’s data, personal settings, and application settings with a core OS instance and a shared generic profile.
Applications and user personality are injected into the core desktop VM, and a brokering mechanism manages connecting the end-users to the VM
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

33. IaaS: Effect of Virtualization and Cloud Computing on Data Centers
Since economies of scale are in play, larger DC is in favor. So the HW blocks can be enumerated.
How do you design for best performance?

34. Summary and Conclusion
Cloud architecture consists of IaaS, PaaS, and SaaS
IaaS is the foundation of Cloud Computing
Virtualization is central technology for IaaS
IaaS provides set of services: Servers, Storage, Network, Integration, Management, and Payment/Billing
Virtualization and Cloud has a significant impact on the Data Centers

35. END