Next Generation System Architecture
Jerry Huck CTO – Business Critical Systems HP Chief Scientist
© 2004 Hewlett-Packard Development Company, L.P. The information contained herein is subject to change without notice
The next big thing
Price + Reliability + Security + Simplicity and manageability + Adaptability + Innovation + Connection
September 22, 2004
Draft
2
HP strategy
To offer a portfolio of products, services and solutions that are high-tech, low-cost and deliver the best customer experience “Innovation at a price our customers can afford, delivered with an experience that sets us apart.”
– Carly Fiorina, HP CEO Focused innovation that drives real value Affordable technology that offers the best return on investment
ch -te
w Lo -co
Hi gh
st
Customer
Best customer experience
The best customer experience with HP and HP
September 22, 2004
Draft
3
The big shifts
All processes and content will be transformed from physical and static to digital, mobile and virtual. The demand for simplicity, manageability and adaptability will change how customers work and organize, buy and use technology. It’s a horizontal, heterogeneous, networked world. Standards are about connection and common language.
September 22, 2004
Draft
4
Mobility
•
Market drivers Demand for new compelling services continues Need new services (data + voice) to offset declining voice revenues Tools to increase mobile productivity Investments in wireless LAN technologies and device
Draft
Consumers
Carriers
•
Business
• •
• September 22, 2004 Application
5
Next Generation Technology Challenges and Opportunities - Itanium Systems - Blades - Virtualization - Storage Grids
September 22, 2004
Draft
6
Itanium Systems – Building an enterprise system with standard processors
September 22, 2004 Draft
7
Intel Itanium Architecture: Designed for Business Critical Next Enterprise Intel Computing ~6 Architecture Itanium
® ®
EPIC
Instr / Cycle ~2
Processor
Age : 2+
• Larger
and more demanding workloads require new approach:
Superscalar Performance
~1 ~.3
• Designed
RISC
HP PA-RISC, Sun SPARC, IBM Power, MIPS, Alpha
Age : 9-15+
CISC
IBM 370, VAX 11
Age : 20+
Time Performance through parallelism
• •
for 64-bits from the ground up • Architected for performance, scalability, and business critical availability
Beating the memory latency gap & shorter Pipeline
• • • • •
Massive on-chip resources
•
Business Critical Availability
•
Built-in instruction-level parallelism Issue ports and execution units support up to twice as many instrs/clock cycle Maturity curve narrows clock speed 22, 2004 September gap over
128 general registers, 128 floating point registers, 8 branch ((vs 16 on x86) Fewer memory accesses (loads/stores) on complex workloads
Very large virtual and physical address spaces Shorter memory pipeline Latency avoidance Predication of instruction execution Data and control speculation
•
Security: sophisticated ring protection and buffer overflow protection Protected data paths Failure mode analysis
• •
•
Draft
Intel Itanium processor family and RoIT
• Open
industry standard platform
− common platform for standard OS strategy − lowers HP’s R&D and support costs − ensure ISV adoption − ride industry standard ROI curve
• Reduces
complexity
− provides legacy continuity − common management reduces IT TCO − common architecture provides flexibility for dynamic utility computing reconfigurations
September 22, 2004
Draft
9
HP’s industry standards-based server strategy Moving to 3 leadership product lines –
Current
HP NonStop server
Mips
built on 2 industry standard architectures Future
Industry standard
HP NonStop server Enabling larger investment in value-add innovation
Itanium®
HP Integrity server
Itanium®
Common technologies
•
HP 9000 / e3000 server HP AlphaServer systems HP ProLiant server
x86 Alpha PA-RISC
HP Integrity server
Itanium®
Adaptive Manageme nt Virtualizatio n HA Storage Clustering
10
•
HP ProLiant server
x86
• • •
September 22, 2004
Itanium® 2 & Xeon™ Processor MP Comparison
Performance Reliability Scalability
Itaniu Xeon™ Characteristic MP m® 2 Error recovery on data bus ECC +30%-50% Internal soft +50%-100% error logic 2005 in ‘04 um i ms n in ’07+ check Ita tfor Machine Check rms Pla atfo Pl d Architecture ase B onLaw l Xe Bad data te ore’s In Mo containment Cache 2005 Reliability ’07+ Lockstep ‘04 * For Enterprise & Technical Computing support SDEC, Application Segments Memory retry double Today: bit 30-50% higher performance Memory spares node Partitioning 2007:
60 0 50 0 40 0 30 0 20 0 10 0 0
+125% +50%
4 TPCC P
3 PTPCC 2
X onM Itanium2 e P
Up to 2x performance at same platform price
High-end “RISC”-level RAS
node
~2x Higher Scalability
Itanium’s EPIC Architecture: Highest Performance, Reliability, Scalability
* Other names and brands may be claimed as nterprise others. E the property of Platforms
Group
Other names and brands may be claimed as the property of others.
11
Itanium-based servers will soon exceed IBM Powerbased servers in unit volume! (2004 or early 2005)
IPF & RISC Server Volumes ( ku)
500 400 300 200 100 0
2002 2003 2004 IPF 2005 2006 Power
Source: IDC
September 22, 2004
2007
Draft
12
•
“Inventing” a dual core Itanium® CPU
The standard Intel cartridge packaging is not at maximum density
− CPU silicon chip is in a package, on a carrier board, with power on the end − The basic chip and package could be packed much more densely
Intel CPU “carrier” board
Industry-standard Itanium2® power pod (DC to DC power conversion)
Intel CPU chip inside package HP external cache and controller chip HP CPU “carrier” board
HP power solution (goes on top rather than on the end)
September 22, 2004
Draft
13
sx1000 chipset error correction features
PCI online add/replace
PCI/PCI-X Buses
32 SDRAM DIMMs
DRAM chip-spare
Cell Board
2X MEM BUFFER 2X MEM BUFFER 2X MEM BUFFER 2X MEM BUFFER
Itanium® 2 Buses
CPU
Socket*
CPU
CPU bus ECC
Socket*
CPU Cache ECC
CPU
CPU
Socket*
PCI-X PCI-X Host … Host Bridge Bridge
16 Links
Socket*
…
Parity on PCI-X and “ropes”
DIMM Address + Control Parity
ECC + single wire correct on M-links
Cell Controller
I/O ECC PCI-X System + single Bus Adapter wire correct
ECC/Parity Internal to VLSI
I/O Subsystem
Crossbar
September 22, 2004
ECC + Single wire correct on crossbar links and VLSI
crossbar hard partitioning
Draft
14
HP Integrity and HP 9000 Server Roadmap
Current offering
Revision 4.5
July-04
2004
2005
2006
4-128P
HP Integrity Superdome
CPU: Itanium2 Madison 9M CPU: Itanium2 Madison9M CPU: Itanium2 Madison9M CPU: Itanium2 Madison9M
DDRII
CPU: Montecito New Chipset CPU: Montecito New Chipset
PCI-E
2-16P 2-32P
HP Integrity rx7620-16, rx862032
DDRII
PCI-E
1-8P
HP Integrity rx4640-8
CPU: Montecito New 8p Server & DDRChipset II PCI-E CPU: Montecito CPU: Montecito New 4p Server & DDRChipset II PCI-E CPU: Montecito
1-4P
HP Integrity rx2600-2
1-2P
HP Integrity rx1600-2
New Chassis Intro.
CPU: DP Itanium
All upgrades “in-box” except as noted
Draft PCI-Express
Timeframes not to scale Plans subject to 2004 September 22,
CPU: Next Gen & Chipset PCI-EDDRII DP Itaniumprocessor release *Not available at initial
DDR-II Memory
15
Near Futures Summary
•
Third cellular chipset planned for Montecito shipment in 2005
− Will also support PA-8900 on this new chipset
•
Increased Performance
− − − − − − 1.3X CPU bus bandwidth increase 2X memory bandwidth increase 4X memory capacity 4X crossbar bandwidth increase 3X aggregate I/O bandwidth increase 2X I/O slot bandwidth increase (PCI-X 2.0 DDR)
•
Increased Reliability and Resiliency
− − − − Double DRAM sparing (tolerate failure of 2 DRAMs on a DIMM) Multi-path crossbar topology with self-healing links Redundant DC to DC converters and system clocks Greater fault protection within the CPU
Draft 16
September 22, 2004
HP server and storage portfolio
The world’s broadest, most robust enterprise offering
Industry standard servers Traditional RISC servers StorageWorks
Application
Database
Integrity Servers
NonStop Servers
Multi-OS Management Services
September 22, 2004 Draft
NonSto p OS
- Tape and optical storage
Access
Alpha servers
ProLiant Servers HP9000 servers
- Storage arrays and NAS - SAN infrastructure, software,media Tru6 4
17
Blades – toward a utility computing fabric
September 22, 2004
Draft
18
HP Server Blade Portfolio
Designed for adaptive, multi-tiered architectures
new
1P Front End
2P Stateless
New
g m in Co 005 2
2P Mid-Tier
4P Back-end
September 22, 2004
Draft
19
HP BladeSystem management:
New suite automates IT service delivery
Optimizing HP Systems Insight Manager and iLO to create the hub of automated, virtualized HP BladeSystem management New plug-in controllers to SIM
1 Integrate 2 monitor/control
Virtualize
• Virtualization controller: Deploy, migrate, and manage heterogeneous virtual nodes • Automation controller: Workflow automation for policy-driven change of physical and logical resource pools
4
-Compute -Storage -Network -Power -Software
3
• Patch and scanning controller: OS and application vulnerability patch • Intelligent Networking: Failover network path for maximum availability and performance
Adapt and Advise and change Automate Automated infrastructure control across the lifecycle
• Dynamic power management: maximize energy efficiency & • Services to accelerate valuebalance power draw and ensure success • Affordable blade management software/hardware bundles
Draft 20
September 22, 2004
Drive Out Hard Costs
Traditional Rack-Mount Servers ProLiant DL360G3 w/ Ethernet and SAN 40x3 network + 40x2 power = 200 cables 36 servers + Ethernet switches 16.6kW* 1+1redundancy ProLiant p-class blades up to 19% savings 5x2 network + 4 power = 14 cables 93% reduction 48 - 96 servers + Ethernet switches 25 - 60% reduction 12.1kW 27% savings Less PDUs (see whitepaper
for whole list)
Acquisition Costs (based on 8 servers) Cabling Connectivity (based on 40 servers) Data Center space (density per 42U rack) Power and Cooling (based on 40 servers) Failover servers
N+1redundancy ~50% savings
September 22, 2004
Draft
21
Virtualization – flexibility and agility in IT
September 22, 2004
Draft
22
servers will dramatically improve server utilization rates, increase server flexibility and reduce the overall spending required for servers…Virtualization should become an ongoing effort and part of the server strategy for every enterprise.”
Gartner Server Virtualization Evolves Rapidly, 2003
Adaptive Enterprise vision
Business and IT synchronized to capitalize on change
1.
Measure, assess and maintain a dynamic link between business and IT
MEASURE & ASSESS time, range, ease
Business processes
EXTEND & LINK
Suppliers Employees Customers
2. 3.
Architect and integrate heterogeneous IT environments Extend and link business processes across suppliers and customers Manage and control business processes, applications and the whole IT environment
Applications
ARCHITECT & INTEGRATE
simplify, standardize, modularize, integrate
MANAGE & CONTROL assess, advise, act
Virtualized resources
4.
Infrastructure
September 22, 2004
DraftUnder Non Disclosure until December 4, 2003
page 24
Adaptive Enterprise design principles
simplification
• • • •
Reduce number of elements Eliminate customization
+
Use standard technologies + standardization interfaces
+
• •
Adopt common enterprise architecture Implement standard processes Break down monolithic structures Create reusable components
Applied consistently across:
• • •
modularity
•
Business processes Applications Infrastructure
+
integration
Implement logical • architectures Manage the dynamic link between business + IT
• •
Connect apps + processes inside and outside
DraftUnder Non Disclosure until December 4, 2003 page 25
September 22, 2004
Virtualization
An approach to IT that pools and shares resources so utilization is optimized and supply automatically meets demand Business
Supply Demand
Information technology
September 22, 2004 DraftUnder Non Disclosure until December 4, 2003 page 26
Just in time delivery changes the economics
Just in case
ERP CRM Silos
Just in time
ERP CRM Web
Server Storage
Web
IT
•Pooled
Network Software •Shared
Manufacturing
Pooled Inventory
Engine Seats Engine Wheels Seats Chassis
page 27
Excess inventory
Wheels Chassis
DraftUnder Non Disclosure until December 4, 2003
September 22, 2004
Optimizing resources from desktop to datacenter
VIRTUALIZATION INNOVATION TODAY Complete IT Utility
Optimize all heterogeneous resources so supply meets demand in real time
Business Value
Integrated Virtualization Element Virtualization
Optimize utilization of server, storage or networking resources
Servers Storage NetworkSoftware
Optimize environments to automatically meet service level agreements
Strategic Importance
September 22, 2004 DraftUnder Non Disclosure until December 4, 2003 page 28
Environment
Flexible Virtualization for HP-UX 11i
Virtualization Techniques
Process Management (single OS with or without psets) Hard and Soft Partitioning (multiple OS images) On Demand High Availability (across partitions or servers) Disaster Tolerant
App A App B
Partition A Partition A Partition
Partition B Partition B Partition
HP Process Resource Manager
HP nPartitions or HP Virtual Partitions
Instant Capacity On Demand (iCOD)
Metrocluster HP Serviceguard & Continentalcluster s
Intelligent Policy Engine: HP-UX Workload Manager (HP-UX WLM) Single Server Groups of Servers
Draft
September 22, 2004
Geographically Dispersed Groups of
29
HP Integrity Virtual Machines … optimum utilization across Multi OS • Sub CPU virtual machines
with shared I/O • Runs on a server or within an nPar • Dynamic resource allocation built-in • Resource guarantees as low as 5% CPU granularity • OS fault and security isolation • Supports all (current and future) HP Integrity servers • Designed for multi OS – first on HP-UX 11i • VSE integration for high availability and utility pricing
app1 app3 app5
app2 app4 app6
app1 app3
app2 app4
app1 app3
app2 app4
OS (Linux)
OS (HP-UX 11i v2) OS (HP-UX 11i v2)
Intelligent Hypervisor
Hardware
Memory I/O I/O
I/O
2H 2005
September 22, 2004 Draft
30
Virtualization requires a new approach
• Demand service levels not dedicated systems
People
• Transform “server huggers” into “service providers” • Measured and reward resource utilization and SLAs • Simplify and standardize IT processes based on best practices so you can replicate and automate • Monitor and charge back IT resources based on actual usage • Pool and share standard, modular IT resources
Process
Technology
• Allocate resources dynamically to ensure SLAs are met • Optimize utilization and availability • Automate flow of resource supply to meet demand
September 22, 2004
DraftUnder Non Disclosure until December 4, 2003
page 31
Grids and Storage – another key part of a utility
September 22, 2004
Draft
32
Evolving storage towards the Adaptive Enterprise
Storage Grid enables NEW information services Network Storage enabled Greater scalability Efficient resource sharing Higher system availability Shorter backup windows
Real-time scalability Scalable capacity
Storage Grid
Network Storage
DAS
Resource efficiency Real-time information services
September 22, 2004 33
HP StorageWorks Grid: A unique “array”
Hostconnected networks
High performance inter-smart cell networks
Storage clients
September 22, 2004 34
HP StorageWorks Grid composition
•
Smart cell hardware
− − − − Processor Cache Internal disks, tapes, etc. Off-the-shelf components Federates with peers Places data on internal devices Provides smart cell “personality” Responds to changes
• Redistributes data and workloads
Host Facing Network
Inter-smart cell Network Intercell IOPs
Host IOPs
Processor, cache
•
Smart cell software
− − − −
Devices
Service modules
3rd party modules
− Hosting environment for 3rd party code
•
Smart cells are federated into domainsCommon operating
software
September 22, 2004
Plug-in environment
35
Value
•
HP StorageWorks Grids can exist on their own
− No non-grid storage is needed − But customers will already have some
•
HP StorageWorks Grids can coexist with conventional storage
− Customers can plug into existing storage networks
•
We expect customers to deploy multiple HP StorageWorks Grids
− each HP StorageWorks Grid may contain multiple domains
•
September 22, 2004
Conventional arrays can be migrated into the grid
36
Conclusions
September 22, 2004
Draft
37
Future System Architecture
•
HP shows proven technology innovation that drives real business results
te ch Hi gh
− Select what’s best for you: open architecture approach across multiple operating environments.
•
Lo w
Strategy that offers the best return on investment
− No need to choose between highperformance computing and industrystandard economics − Unsurpassed agility and proven superior price/performance
Customer
Best Customer experience
co st
•
Dedication to customer satisfaction that creates a superior customer experience
− Collaborative customer relationship programs, enabling open and frequent communication − Clear roadmaps and commitments September 22, 2004
38
Contact Info
−Jerry Huck CTO, HP Business Critical Servers Division.
[email protected] , Tel:+1408-447-2429 −Harel Ifhar. HP Global Alliance Manager for Amdocs.
[email protected] Tel:+972-524840-916 −Esteban Birenbaum. HP Technical Alliance Manager for Amdocs.
[email protected] Tel:+972-52-4840918
te ch gh Hi Lo w st co
Customer
Best Customer experience
September 22, 2004
Draft
39