© 2009 wipro ltd - confidential private cloud at wipro cloud computing based on condor
TRANSCRIPT
© 2009 Wipro Ltd - Confidential
Private Cloud at Wipro
Cloud computing based on Condor
© 2009 Wipro Ltd - Confidential2 © 2009 Wipro Ltd - Confidential2
3
1 Background
2 Wipro Private Cloud
System architecture
4
Agenda
© 2009 Wipro Ltd - Confidential3 © 2009 Wipro Ltd - Confidential3
Need:
• Share physical infrastructure between multiple projects and CoE's (Center of Excellence) to reduce server sprawl and number of physical labs
• Provide an environment for evaluating new technologies, developing solutions and enabling collaboration between multiple labs
• Centralize infrastructure procurement and management• Reduce infrastructure cost of CoE's by enabling multiple
development environments
Solution:
• Setup a private cloud for virtual compute and application infrastructure
• Build a self-service portal for on-demand provisioning to reduce process overheads
• Support multiple types of virtualization software• Reuse existing physical infrastructure, procure minimal new
infrastructure
Background
© 2009 Wipro Ltd - Confidential4
Wipro Private Cloud
© 2009 Wipro Ltd - Confidential5 © 2009 Wipro Ltd - Confidential5
SaaS Mgmt
Wipro Users
SaaSUser Intranet
SaaSApp
SaaS Enablers
Wipro Cloud Portal / Web Services API Layer
Physical Resource Pool - Servers, Storage, Network
Managed Network
CloudAdminC
loud
O
A&
M P
ortal
Wipro Cloud Core• Virtual Machines • Shared Storage• Virtual Appliances• Application Services
• Automated Provisioning• Multi-tenancy & Isolation• Cloud Accounting & Auditing• Performance & Fault Monitoring• Automated Network & Security
Developers
Virtual Lab
Managed Network
Wipro Private Cloud
© 2009 Wipro Ltd - Confidential6 © 2009 Wipro Ltd - Confidential6
Service Element Service Feature
Compute Servers
Virtual desktop – equivalent to 1.2GHz,512MB RAM,10GB HDD,25Mbps N/w
Low End Server – equivalent to 2x1.2GHz, 2GB RAM, 20GB HDD, 25Mbps N/w
High End Server – equivalent to 4x1.2GHz, 4GB RAM, 40GB HDD, 25Mbps N/w
OS types Linux (CentOS, RHEL) and Windows XP/Server on Intel x86, x86_64 architecture
StorageISCSI (RAID 5), NFS and CIFS
Data persistence across power-off, suspend & resume of VM’s
Public images/ appliances
Ready-to-use public images RHEL 5, Windows XP, LAMP (CentOS 5.2, Apache, Axis, Tomcat,
MySQL, PHP, Python)
Preconfigured Software load balancer, firewall appliances
NetworkIsolation between CoE's resourcesIPSec, SSL based VPNPublic and Private IP Addresses with NAT support
Private images Can Upload VMware Server, VMware ESX and Xen Virtual Machine Image formats
Reports Reporting on CPU, Storage and memory usage back to user
Cloud Services catalogue
© 2009 Wipro Ltd - Confidential7 © 2009 Wipro Ltd - Confidential7
• L1 - Virtual Servers on demand• Virtual servers, desktops, storage• Migration assistance• Self-service portal
• L2 – Application infrastructure on demand• Appliances of standard software• Managed backup, proactive monitoring and help-desk• Itemized billing and charge-back
• L3 – Business service infrastructure on demand• Scalable business services• Multi-tenant application infrastructure (content
management, identity management, database, load balancer, firewall, ...)
Levels of Service
© 2009 Wipro Ltd - Confidential8
System Architecture
© 2009 Wipro Ltd - Confidential9 © 2009 Wipro Ltd - Confidential9
ServiceLayer
LoadBalancer
AppLayer
VirtualMachineLayer
Bare-metalLayer
Mon
itor
ing
Clo
ud M
anag
emen
t
Provisioning•Resource mgmt•Workload mgmt•Auto recovery•Task & Process Automation
Service
LB - Active LB - passive
Inst 1 Inst 2 Inst n
VM 1 VM 2 VM n
Pro
visi
onin
g
Monitoring•Performance•Availability•Alarms•Billing
Management•Service Governor•Policy enforcement•Incident mgmt•Optimizer•Contention
Network Control Alarms
OA&M Portal &Web Service
Gateway
CustomerOA & M Access
•Design, Test•Package, Deploy
Bare-metal design•Standardize•Automate•Re-provisioning
Virtual M/c design•Standardize•Automate•Agile•Caching•Appliances
Service design•SLA’s, Policies, rules, priorities•Packaging•Custom agents•Shared Services•Billing parameters
•Configuration &Change mgmt
Developers Business Users Operations
Private Cloud – in Action
© 2009 Wipro Ltd - Confidential10 © 2009 Wipro Ltd - Confidential10
Architecture & Service layers
Cloud service
© 2009 Wipro Ltd - Confidential11 © 2009 Wipro Ltd - Confidential11
Identity Management
Identity Management
Bare-metal provisioningBare-metal provisioning
Storage provisioningStorage provisioning
N/W provisioningN/W provisioning
VM provisioningVM provisioning
N/W Plug-inN/W Plug-in Storage Plug-inStorage Plug-in
Bare-metal Plug-inBare-metal Plug-in VM Plug-inVM Plug-in
Grid SchedulerGrid Scheduler
Service GovernorService Governor
Customer Portal
Customer Portal
Workflow ManagerWorkflow Manager Cloud State
VM CachingVM Caching
Web Service Gateway
Web Service Gateway
VM Repo
Legend: Developed in WiproDeveloped in Wipro 3rd Party components
3rd Party components
Charge-backCharge-back
N/W (nagios) Monitoring
N/W (nagios) Monitoring
Nagios plug-inNagios plug-in
AlertsAlerts
Metrics MonitorMetrics Monitor
In DevelopmentIn Development
System Components
© 2009 Wipro Ltd - Confidential12 © 2009 Wipro Ltd - Confidential12
Cloud Backbone10.201.72.0/24
Cloud Mgmt192.168.3.0/24
Project X192.168.5.0/24
Project Y192.168.6.0/24
Project Z192.168.7.0/24
Mgmt ServerHA Pair
Router, firewallVPN Server, IPS, IDS, NAT
Switch FabricCloud physical systems
Storage
VM
VM
VM
VirtualMachines
VirtualStorage
Isolated network per
project
Deployment Example
© 2009 Wipro Ltd - Confidential13
Use of Condor
© 2009 Wipro Ltd - Confidential14 © 2009 Wipro Ltd - Confidential14
• Trusty old features– Flexibility – ClassAd mechanism, configurations and
policies– Web Services API– High availability– Resource utilization of jobs
• Newer features we like– VM Universe– Partitionable Slots– Lease management– Integration with Amazon EC2 (public cloud)
• Proven in large scale deployments• Condor-users and condor-admin support• Open source
Why Condor?
© 2009 Wipro Ltd - Confidential15 © 2009 Wipro Ltd - Confidential15
• Mostly standard configuration• A few custom class ads in jobs and machines• Schedd and Collector configured in HA mode• Condor spool for VM persistence• Virtual machine provision request handled by
Condor– VM job to physical machine match-making, file transfer
• Partitionable slots for dynamic partitioning of physical machine resources
• Customized condor_vm_* files for configuring and starting VM's
– VLAN control, Swap disk and additional storage creation, ...
• Lease management for limiting the number of running instances of a licensed image
How are we using Condor?
© 2009 Wipro Ltd - Confidential16 © 2009 Wipro Ltd - Confidential16
Working with Condor:– With advanced Condor skills, a lot can be achieved without
modifying condor code
Workarounds:– Passing number of virtual CPUs to VMware– Patch to pass proxy username and password to gSOAP for
EC2 integration– Patch to get VM resource usage details on ESX– Special configuration to handle 2 hour delay in detecting a
few execute node failures (Thanks Todd!)
Feature wish list:– Remote IWD support for VM universe, to avoid any file
transfer– Live migration of VM jobs
Observations, Workarounds, Wish list