Dr. Kamal Bhattacharya IBM Research - India Cloud Computing
Overview From Enterprise Data Center Consolidation to Cloud Computing A Definition IT Management and Cloud Migration to Cloud Future of Cloud 2
A (fiscal) quarter in the life of a Data Center Peak 100% Avg 33% Peak 56.4% Avg 6.4% Peak 13.4% Avg 0.1% Treemap Legend: Boxes are servers categorized by type Size of a box indicates peak utilization Color indicates average utilization Low utilization of servers High variability of /MW Highly customized, few standards
Server Virtualization Run many physical systems as virtual machines on one physical hardware ( like-to-like ) Applicatio Applicatio n n Application Hardware Hardware Hardware Image Migration Application 1 Virtual Hardware Application 1 Virtual Hardware Virtual Machine Monitor Hardware Application 2 Virtual Hardware
Virtual Machine Monitor (VMM) Example KVM Full Virtualization Application 1 User Space (apps) Guest (Virtual Machine) Single Linux Process QEMU I/O Virtual Hardware VMM Hardware memory /dev/kvm Memory VMM (Hypervisor) Linux Kernel I/O CPU Sufficient criteria for VMM s (Popek, Goldberg CACM July 1974) Equivalence / Fidelity A program running under the VMM should exhibit a behavior essentially identical to that demonstrated when running on an equivalent machine directly. Resource control / Safety The VMM must be in complete control of the virtualized resources. Efficiency / Performance A statistically dominant fraction of machine instructions must be executed without VMM intervention Papers: Adams, Agesen A Comparison of Software and Hardware Techniques for x86, ASPL 2006 Bugnion, et.al, Disco: Running Commodity Operating Systems on Scalable Multiprocessors, ACM 5 Transactions on Computer Systems, November 1997
Workload Mobility User Space (apps) Guest (Virtual Machine) /dev/kvm QEMU VMM (Hypervisor) Linux Kernel User Space (apps) Guest (Virtual Machine) /dev/kvm QEMU VMM (Hypervisor) Linux Kernel Memory I/O Memory I/O CPU CPU SAN Storage 6
Image versus Instance Instances RHEL 5 Image RHEL RHEL RHEL 5 /dev/kvm QEMU VMM (Hypervisor) Linux Kernel Memory CPU I/O Parent RHEL 5 RHEL RHEL RHEL Child1 Child2 Child3 7
A virtual resource exposed as a Service 8
Virtual Appliances Provisioning of resource instances based on standard resource representations develop capture request Image Provisioning deploy Library Manager Virtual Appliance Instance Hardware Virtual Appliance VMM (Hypervisor) Linux Kernel Physical Hardware Virtual Appliances (or Images) are packaging of virtual representations with reference to key components that enable the running instance For example A linux virtual appliance is a read-only file-system that contains everything required to run (at least a part of) the service Does not contain the kernel, only a reference to the kernel Does contain a manifest describing the appliance Examples: AMI (Amazon Machine Image), OVF (Open Virtualization Format) 9
Workload Profiles #requests Provision only what you need Scale up to peak as required Scale down to plateau and free up resources 1Q 2Q 10
Operational Styles Your Data Center Your Data Center Virtualized Somebody else s Data Center Pay for servers & maintenance Pay less for servers & maintenance Rent servers, pay by usage 11
A Cloud computing model provides u Internet-based access to virtualized resources u Elastic scaling of virtualized resources u Utility-based pricing model 12
Virtualization abstracts the behavior of a system from its actual physical implementation A virtualized resource behaves like the physical resource it emulates by leveraging the capabilities of the physical system such that it provides: fidelity ( feels like the real thing ) containment ( no conflicts with other virtual resources ) performance ( performs almost like the real thing ) 13
A System that implements a Cloud Computing Model provides a Service that provides access to virtualized resources over the Internet provisions and de-provisions virtualized resources as required charges only based on actual consumption Here is the challenge: A Service is a means of delivering value to a client by facilitating outcomes the client wants to achieve without the ownership of associated costs & risks 14
Architectural Styles Higher switching Costs Dedicated Infrastructure as a Service (IaaS) Platform as a Service (PaaS) Software as a Service (SaaS) Application Data Middleware Storage Network Application Application Data Data Data Middleware Middleware VMM Storage Network Application Application Application Data Data Data Middleware VMM Storage Network Application Data Middleware VMM Storage Network Amazon Microsoft Azure, Google Apps SalesForce.com Higher degree of standardization 15 IBM confidential
Operational Styles Enterprise Data Center Managed Service Provider Public Cloud Client 1 Client 2 Client 3 Private Cloud Shared Private Cloud Consolidated IT Infrastructure Capital expense reduction Some operational efficiencies Hosted IT Capital expense reduction Outsourced service management with contractual SLA s Hosted IT Capital expense reduction Utility based costs Best-effort SLA s 16
Implications of the Cloud Computing Model Do we need a new programming model? Do we need new IT management processes? How to move to Cloud? 17
Cell: d02wdm001network d02wdm001.southbury.ibm.com App: tools_erdm Module: ERDM.war App: tools_drms Module: drms.war App: procurement_sqms Module: sqms.war App: learning_navigator_admin Module: LearningActivityWS.war Module: LearningAdminTool.war App: hr_idp Module: IdpWeb30.war App: hr_exec_comp_personal Module: hr_exec_comp_personal_web.war App: financing_tools_ifs Module: financing_tools_ifswa.war App: finance_treasury_emea_tps Module: tps.war App: finance_tools_tmf.ear Module: finance_tools_tmf.war App: finance_tools_lenovotmf Module: finance_tools_lenovotmf.war App: finance_tools_ledgerreporting Module: fiwlrbrio.war App: finance_tools_feat Module: finance_tools_feat.war App: chq_legal_lfms Module: LFMS.war App: EPMEAR Module: EPM.war Cell: Unknown d02was048.southbury.ibm.com DB2 Instance: caeadmin Database: IDPSES Database: HR_IDEAS Database: EUHADBM0 Database: DRMS Database: DPESREQ3 Database: DPESREQ1 Database: DPESCDA Database: DP1HS Database: DHRAP830 Database: DD1H Database: CABECP Database: DPESHUB1 Database: APUSESS1 Database: APNBRIO Database: APSBRIO Database: XDSPROD Database: TRADE3DB Database: TMFDB Database: SIMSAIX Database: SESSIONS Database: SESSION Database: DP1H Database: PDBPROD Database: NRAADB2D Database: LNPDB Database: LERDP2S Database: R0ADB2 Database: PORTSES Database: PORTALS Database: PORTALP Database: IFSW3SES Database: IFSSTAGE Database: CASHWAS Database: TMFDP1H Database: CASHPP Database: TMFLDB Cell: Unknown Database: IDPSES Database: HR_IDEAS Database: EUHADBM0 Database: DRMS Database: DPESREQ3 Database: DPESREQ1 Database: DPESCDA Database: DCSDP1HS Database: DHRAP830 Database: DCSDD1H Database: CABECP Database: APUSESS1 Database: BRIOREP Database: METRICSX Database: XDSPROD Database: TRADE3DB Database: TMFDB Database: SIMSAIX Database: SESSIONS Database: SESSION Database: DCSDP1H Database: ESKBXXX Database: EODCTL Database: PDB Database: NRAADB2D Database: LNPDB Database: LERDP2S Database: R0ADB2 Database: PORTSES Database: PORTALS Database: PORTALP Database: DCACHE Database: IFSW3SES Database: IFSSTAGE Database: CASHWAS Database: CASHPP Database: TMFLDB Distributed Applications Pictures from discovery data, courtesy: Hari Ramaswamy, Nicolai Joukov, Birgit Pfitzmann 18
Scalability of Distributed Applications Web Server App Server DB Consistency: Each client always has the same view of the data Availability: Each client can always read & write Partition Tolerance: System works well across network partitions CAP Theorem: You can only pick two! Maintaining consistency is a strength of RDBMS, but partition tolerance is not New models (Amazon S3, nosql solutions) prefer to focus on availability and partition tolerance with introducing a notion of eventually consistent Papers: W. Vogels. Eventually Consistent. ACM Queue vol. 6, no. 6, December 2008 Blog: http://www.julianbrowne.com/ article/viewer/brewers-cap-theorem 19 PaaS & SaaS offerings deal with these issues internally
So far we ve been talking about Capex benefits. How about Opex? 20 Source: Forrester 2008
IT Management Challenges Service Catalogue Management Service Level Management Risk Management Capacity Management Availability Management IT Service Continuity Management Information Security Management Compliance Management IT Architecture Management Supplier Management Service Asset and Configuration Management Service Validation and Testing Evaluation Release Management Change Management Knowledge Management Event Management Incident Management Problem Management Request Fulfilment Access Management Is expected??? Can it still apply at the lowest possible cost point? 21
IT Management Segmentation Ticket Application Data Add functional component SLA A Performance Service Provider A Design & Provision Middleware Configure data source Add UID/pwd SLA B VMM patch Service Provider B Storage Backup/restore Network Add port Service Provider C Cloud forces to rethink IT management strategies and may enable a vibrant (complex) eco-system of service providers 22
Cloud Reference Architecture (IBM) Cloud Service Consumer Cloud Service Provider Cloud Service Creator Cloud Services Common Cloud Management Platform Service Manager Business Manager Cloud Service Integration Tools Existing & 3 rd party services, Partner Ecosystems BPaaS SaaS Service Consumer Portal & API S Operational Support Services Service Automation Management Service Request Management Provisioning Service Delivery Catalog Change & Configuration Management Incident & Problem Management Image Lifecycle Management IT Service Level Management BSS Business Support Services Customer Account Management Contracts & Agreement Management Subscription Management Service Offering Catalog Service Request Management Pricing Service Offering Management Order Management Entitlement Management Service Development Portal & API Service Creation Tools Consumer In-house IT PaaS Monitoring & Event Management IT Asset & License Management Capacity & Performance Management Platform & Virtualization Management Metering Rating Billing Clearing & Settlement Accounts Payable Accounts Receivable IaaS Service Provider Portal & API Deployment Architect Transition Manager Operations Manager Security & Risk Manager Customer Care Inf rastructure 23 Security, Resiliency, Performance & Consumability Governance
Moving to the Cloud: ROI? 24
Migration to the Cloud App App App MW MW MW MW Re-Install Cost & Risk App MW MW App MW App MW Migration to PaaS Major MW upgrade App App App App MW MW MW MW Virtualize & Package Client Landing Zone Prod 25
Use Case 1 (Basic Image Migration) Each shape indicates a specific type of system (e.g. an ) The source environment has many different systems, the target only allows two types of systems Migration Client Environment Target Environment
Use Case 2 (Image Migration & Standardization) The target environment allows only two types of systems The target requires the systems to meet certain criteria (e.g. an installed set of tools) prior to on-boarding Migration Client Environment Target Environment
Use Case 3 (Image adjustment & standardization) The source environment has management devices installed that are not commensurate with the target devices required Migration Client Environment Target Environment
Use Case 4 (Application Migration) The arrows indicate sub-system dependencies The target is agnostic to the sub-system and will support recreation of sub-system dependencies Migration Client Environment Target Environment
Use Case 5 (Migration to a standardized Platform) The arrows indicate sub-system dependencies The target requires a specific type of sub-system hence requires the source sub-system to be migrated to the target sub-system Migration Client Environment Target Environment
The Solution? Migration to Cloud?
Let s break it down: Scenario 1 1.1 Transform Migrate 1.2 Transform Migrate 1.3 Transform Migrate
Let s break it down: Scenario 2 1.1 2.1 Transform Transform Migrate 2.1 Transform Migrate 2.2 Transform Migrate
Let s break it down: Scenario 3 A 2.1 A B C 2.2 1.2 + 2.2 B C 3.1 Transform A C B Migrate D D 1.1 + 2.1 D
Summary Cloud Computing is a virtualization based business model It bears various many opportunities and also challenges It is reinforcing rigorous architectural thinking It is opening up new business models and provider eco-systems 35