enabling the autonomic data center with a smart bare-metal
TRANSCRIPT
Enabling the Autonomic Data Center with a Smart
Bare-Metal Server PlatformArzhan Kinzhalin, Rodolfo Kohn, Ricardo Morin, David Lombard
Software and Services Group6th International Conference on Autonomic Computing
Barcelona, SpainJune 17, 2009
2
Software and Services Group
2
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Agenda
•Motivation
•The requirements
•The solution
•The value for autonomic computing
•Summary
3
Software and Services Group
3
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
The Motivation
•Data Centers consist of 10s or even 100s thousands of commodity servers−Modern applications that Data Centers run are designed to scale-
out and thus require dynamic allocation of the resources
•The Data Centers employ management software to discover, query, provision, configure, allocate and de-allocate resources
•Nevertheless, there is an automation gap which is referred to as time-zero problem
4
Software and Services Group
4
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Time-zero Problem Illustrated
4
A general-purpose BIOS forces clusters, enterprise grid, or cloud, which have well-defined but limited operational modes, to be operated as PCs in racks, instead of a robust, scalable, integrated entity
vs.
5
Software and Services Group
5
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Time-zero Problem and PXE
•PXE is widely used to address the time-zero problem
•It allows booting arbitrary OS image−Normal management layer takes it from there
−Management tools are proprietary
•PXE is static−Bound to MAC, normally configured manually
−Dynamic provisioning capabilities are minimal
•PXE is unreliable and not scalable−Uses UDP and TFTP
6
Software and Services Group
6
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Understanding the Industry Needs
Data Centers need a platform that is
•Based on standards−DMTF, Networking
•Scalable−Routable protocols, low-overhead
•Reliable−Discovery and transport
•Secure−Authentication and authorization
•Available at time zero−Take it off the box, plug it in, turn it on
7
Software and Services Group
7
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Proposed solution
Use the technologies available with modern Intel platforms to expose the server as an manageable entity thus enabling the intelligent hardware and software configuration, provisioning, and management
8
Software and Services Group
8
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Data Center Model
Simplified node roles
•Resource Manager owns and manages the datacenter server repository
•Directory Agent is the SLP DA
•Server Configuration Manager is the policy enforcement server
•Bare-metal servers are the commodity server units
9
Software and Services Group
9
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
The architecture
Standardized Runtime Environment
WBEM interfaces for discovery and access
Capability Inventory
Deployment
Infrastructure Elements
EFI BIOS
Hardware
EFI-bootable OS (transient) or Domain0 (persistent)
Power Management
Firmware Tools API
FWT Driver
Platform configuration Better BootFirmware
Management
Development
Toolkit
Compiler, binutils
Standardizedruntime
Developer’s docs
PXE++
Virtual Machine Monitor (present only in persistent mode)
Deploymenttools
Custom/AdditionalCapability
FWT Driver
ME Firmware
ME
NPTM
Workload Characterization
10
Software and Services Group
10
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
The features
•Runs at time zero on bare-metal platform−All the features are readily available out-of-the-box on power on
•Does not require human intervention once it’s plugged in−Automatic discovery, configuration, and provisioning
•Uses reliable protocols−Both application- and transport-level
•Extensible−Based on standard CIM model
−Easy to develop and deploy custom providers
•Leverages modern DC infrastructure−Uses stable, widely accepted technologies
−Enables smart policy-based resource management
11
Software and Services Group
11
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Technology Overview
WBEM-compliant set of protocols and technologies
•Service Location Protocol (SLP)−RFC 2608
−Reliable discovery protocol
•Common Information Model (CIM) Schema−DMTF standard representation of manageable resources
−It has own XML representation (CIM-XML) as well as bindings to WS-Management
−Extensible
•Security−SSL/TLS transport level security
12
Software and Services Group
12
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Patagonia Lake: Proof-of-Concept Implementation
SLP Directory Agent
WBEM ResourceManager
(developed by any ISV)
2-Subscribe for WBEM services
7-Notification of new WBEM services
1- Subscribes to SLP DA to receive
Notifications of new WBEM services
Configures the discovered server in pre-boot
environment and tells the server to continue booting
3 – Upon initiation, EFI BIOS passes control to Patagonia Lake in preboot. SLP SA and SFCBD
with special providers are running
SLP DA receives WBEM service registration and notifies all registered
applications
11-Continues booting the existing OS (or the
provisioned one)
13
Software and Services Group
13
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Message Sequence Diagram
Manageable Server Directory AgentResource Manager DHCP Server
Request IP address
Assign IP address
SLP: register WBEM service
Subscribe to new registration events
Notify on new service registration
CIM-XML: request server configuration
CIM-XML: configure CPU knobs
CIM-XML: success
CIM-XML: report server configuration
CIM-XML: boot to the production OS
CPUconfiguration
provider
Boot controlprovider
Start CIMbroker
CIM-XML: success
14
Software and Services Group
14
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Proof-of-Concept Overview
•CIM Broker−SBLIM-SFCB, a light-weight low-footprint implementation
•CIM providers−CIM_Processor extension to expose CPU configuration
>Adjacent Sector pre-fetcher, aka Second Sector pre-fetcher
>Hardware pre-fetcher
−CIM_BootControl>Controls the boot sequence
•SLP−OpenSLP, client integrated into SFCB and stand-alone SA
•Linux* kernel + uClibc + busybox for the runtime
All in 1.28MB!
15
Software and Services Group
15
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
The Enabling Technologies
•Intel® Rapid Boot Toolkit−UEFI-compliant BIOS
>Certain features irrelevant to server market removed
>E.g. video, UI, waiting for user input
>Freed up space could be used to place payloads
−Payloads>These are EFI applications
>One of them is Linux*
>IRBT 1.0 leaves 1.28MB for the payload
•kexec mechanism−Replaces kernel with another one
−We used it for fast-boot>No hard reset required
16
Software and Services Group
16
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Enabling Autonomous Data Center
•The runtime environment presented is the first step and one of the possible enabling technologies for the future smart hardware platforms
•The solution enables intelligent Resource Managers that−Discover newly plugged servers
−Creates capability inventory of the server
−Make intelligent allocation and provisioning decisions
•There are many applications>Dynamic inventory
>Power-reduction mechanisms
>Hardware and software configuration
>Provisioning
>Reliable network boot
17
Software and Services Group
17
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Example application: rack-level power management
1
Ethernet/IP, Serial, (other options?)
Moblin
Menlow
Runtime/Libraries
DMTF CIM
Security WS Mgmt
SLP Discovery
CIM Broker
WS EventsSecurity
Field-level CIM Providers
RC detailRC (Menlow…)0-U, tucked away in top, side, or bottom of rack
Display showsreal-time
power utilitzation
Headnode (TBD)
Message Fabric (IB)
Intel Smart Platform
≥12
Mgt Fabric(Ethernet)
18
Software and Services Group
18
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Future work
•Larger flash spaces will enable new features
•Security−WS-Security
−Transport
•WS-Management−via WS-CIM binding
•Persistent VMM
•Production deployments−HPC for starters
19
Software and Services Group
19
6th IEEE International Conference on Autonomic Computing
Barcelona, Spain June 17, 2009 Intel and the Intel logo are trademarks or registered trademarks of Intel Corporation or its subsidiaries in the United States or other countries.*Other brands and names are the property of their respective owners.
Conclusion
•We present a smart server platform which enables extensible representation of the server identity and can be used as the vehicle for autonomic computing use cases
•It provides a better solution for pre-boot environment than the existing ones
•The proof-of-concept demonstrates how CPU configuration can be made scalable and driven by policies