Toward a Liquid Infrastructure - VMwaredownload3.vmware.com/vmworld/2005/sln942.pdf · Moves...
Transcript of Toward a Liquid Infrastructure - VMwaredownload3.vmware.com/vmworld/2005/sln942.pdf · Moves...
Toward a Liquid InfrastructureEnabling Enterprise Java™ Virtualization
Joakim DahlstedtCTO, Java Runtime Products Group
BEA Systems
Agenda
IntroductionCurrent problems for the software infrastructureSolution: Liquid infrastructure BEA liquid infrastructure architectureSome concluding words
About the Speaker
Joakim DahlstedtCTO, Java Runtime Products GroupBEA Systems
BackgroundService oriented architecture
Service Level Agreements (SLAs)Resource management
Virtualization trendsResource poolsUtility computing
Clustering advancementsHigh-availabilityCachingGrid-like computing
Project bare metal
Agenda IntroductionCurrent problems for the software infrastructureSolution: Liquid infrastructure BEA liquid infrastructure architectureSome concluding words
One Server: One ApplicationHigh Redundancy
Unpredictable resource requirementsSecurity concerns
Backup ServersHigh Redundancy
Dedicated backup server / applicationHot standby
App 1 App 2 App 3
Backup 1 Backup 2 Backup 3
CPUs Continually More PowerfulHigh Redundancy
Single CPUs getting more powerfulCPU-Architectures going multi-coreCPU-utilization is decreasing
0
500
1000
1500
2000
2500
3000
3500
4000
Mhz
2005 200620042003
UtilizationAvailable Power
Cluster and Policy Mgmt.: ExpensiveHigh Redundancy and Maintenance Cost
Manual cluster managementAre the service levels met?If they are not, what do you do?
Increase heap-space?Add more CPUs?Start new servers?
Resource Metering: How?High Redundancy and Maintenance Cost
Hard to measure utilization for Java Attribution to different layers non-existent
Capacity planning: How?How much do we need? We don’t even know how much we use!
Utility pricing: How?
Agenda IntroductionCurrent Problems for the software infrastructureSolution: Liquid infrastructure BEA liquid infrastructure architectureSome concluding words
Changes in Two EndsUnited in the Middle
From the top Metering and monitoring high-level SLAs
From the bottomIntegrate and extend virtualization
Micro to Macro ManagementMoves Control from Dev to Admin
Natural Evolution of the BEA InfrastructureExact details to general policies
Today: control the detailsTomorrow: set policies
Service Level Based PoliciesRequired # of operations / minAverage latency of operationMaximum latency of operationHigh availability requirements
Two instances on different machines
Resize the Cluster AutomaticallySimplifies SLA Management
Cluster Manager monitors provided SLAsCooperation between WLS and JRockit to make the right decisions
Add/remove power to/from a running instanceAdd/remove nodes to/from the clusterBuilt-in diagnostics to find
Reduces the costs of server management
Partition a Box into Many VirtualIncreases Utilization of a Single Box
Take advantage of VMware to:Divide a physical box into multiple boxesTo isolate and guarantee resources
Optimize ExecutionJava Optimized For VMware VMM
Adding a VMM worsens performanceNormally many OS-JVM conflictsThe two above together =Teamwork (JVM/VMM) to optimize:
Raw speed / pausetimes Suspend / resume / migrateMemory footprint
Create a JVM Resource PoolRemoves Resource Boundaries
Dedicated layer tracks machine resourcesThe stack uses this layer to get new JVMsAllocate resources on demandTrack resource mapping & utilization
Map JVMs to serversResource utilization per JVM
JVM Resource Pool
JVM Resource Pool
Service Level AwarenessAll layers have requirements
WebLogic cluster managerWebLogic serverResource brokerJRockit / bare metalVMware VMM
Different levels of detailBusiness operationsWebLogic server requestsPowerUnitsCPUs
SLA
W-SLA
VMRA
VMRA
RA
Local Global
Insert Metering At All LevelsEnables Utility Computing
Already partitioned, isolated and managed: Now measure!Focus on SLAs and resource agreementsEnable in depth service level monitoring Enable detailed resource meteringCommunicate information across the stackMake local decisions locally
Agenda
IntroductionCurrent problems for the software infrastructureSolution: Liquid infrastructure BEA Liquid infrastructure architectureSome concluding words
Liquid Infrastructure ArchitectureVirtualization Enabling
Poolingand
Brokering
Automation
SLA ManagementPolicy and Monitoring
BEA WebLogicServers
BEA WebLogic Cluster ManagerBEA WebLogic Cluster Manager
BEA Liquid Infrastructure Command Center
Tomcat
Geronimo
BEA JRockit JRockit JRockitJRockit
Tomcat
GeronimoBEA WLS
VMware
server H/W
virtual server
virtual server
virtual server
VMware
server H/W
virtual server
virtual server
VMware targetedoptimizations
BEA Java Resource BrokerBEA Java Resource Broker
BEA Bare MetalBEA Bare Metal BMBMBMBM BMBM
WebLogic Cluster ManagerPolicy-based automated cluster resizingAutomates the policies set up by virtual command and controlPolicy-based grid-functionalityRequests diagnostics from underlying layers
Why are the policies not being metHow is each instance doing locally
Request more resources from underlying layersChange WLS request requirementsMore instances
Java Resource Broker
The Java Resource Broker keeps track of the machines that are available for Java executionMonitors and controls resource utilization for each Java applicationNegotiates with the global resource manager for more power
(
Machine
HypervisorOS
JavaProxy
RB Agent (RBA)
RB Agent (RBA)Java
ProxyJava
Proxy
Java Resource Broker
Machine
HypervisorOS
VMM Optimized Java ExecutionOptional componentStandard VMM comes at some cost
Double virtualization: Managed runtimes already virtualizated (JVM)
VMM-optimized Java executionVMware VMM and JVM awareImprove the efficiency of VMM and JVM operations
Remove OS/JVM conflicts Enable high-end high availablity
Virtual Bare Metal
The execution environment for Java on a VMMOptimized for a single Java process
Launched from the standard OSThird-party native libraries on the standard OSOriginal OS file system
JRockit
Server Machine
os
JRockitJavaProxy
JavaProxy
Hypervisor
Bare MetalBare Metal
Agenda
IntroductionCurrent problems for the software infrastructureSolution: Liquid infrastructure BEA Liquid infrastructure architectureSome concluding words
In SummaryBEA’s liquid infrastructure: a natural evolution Improves resource utilization Simplifies systems’ managementIntegrates and extends existing virtualization solutions Shift from micro to macro(SLA) managementSmart teamwork across entire stackEnables utility computing for enterprise Java