Integrity NonStop Server Update - Hewlett...
Transcript of Integrity NonStop Server Update - Hewlett...
© 2006 Hewlett-Packard Development Company, L.P.The information contained herein is subject to change without notice
Integrity NonStop Integrity NonStop Integrity NonStop Integrity NonStop Server Update:Server Update:Server Update:Server Update:
July 20, 2006July 20, 2006July 20, 2006July 20, 2006
Mary Ahrens, Product Manager, NS14000
Nomi Trapnell, Product Manager, NS1000
NonStop Enterprise Division, HP
SLIDE 219 July 2006
Agenda
• NS14000 Overview:
− NSAA Architecture
− Product Characteristics
• NS1000 Overview:
− NSVA Architecture
− Product Characteristics
• Architecture Availability Comparisons
• NonStop Server Roadmap
• Q/A
Nomi Trapnell
Mary Ahrens
SLIDE 319 July 2006
NonStoparchitecture
NonStop ValueArchitecture
(NSVA)
NonStop Advanced Architecture (NSAA)
Intel Itanium 2 processor based
NS14000 serverAvailable
today
NS16000 serverAvailable
today
NS1000 server Available
today
MIPS processor based
NonStop S-series family
S88000S78000 servers
Available today
NonStop software (with industry-standard APIs)
NonStop infrastructure
HP Integrity NonStop server family
Common software across a choice of hardware configurations
NonStop server offerings
SLIDE 419 July 2006
Integrity NonStop server momentum
• 30% of our business from
Integrity NonStop
• Over 250 partner solutions
available
− 90% of NonStop partners
have Integrity NonStop
solutions
• Deployed in every region
and every major vertical
“We’re looking forward to being on the Itanium growth path.”James Krause, managing director and CIO,Chicago Mercantile Exchange
“It’s important to keep the underlying microprocessor on an aggressive technology roadmap.”Ron Cook, vice president of Technical Strategy & Operations, Radio Shack
“The Integrity NonStop server clearly takes NonStop systems performance to a new level.”Helge Handen, WM-data Card Solutions manager
“We are very pleased that the NonStop server has moved to the industry-standard Intel Itanium 2 processor platform, with its aggressive technology roadmap and significant cost benefits.”Richard Tims, CIO, Electronic Transaction Services Ltd.
SLIDE 519 July 2006
HP Integrity NonStop Servers Architecture
• DMR: Dual Modular Redundancy• TMR: Triple Modular Redundancy
(Availability: seven 9’s)• Itanium2 Servers in a cluster of uni-
processors • Loose Synchronization.
− Each server runs on its own clock.
− Each can perform soft error corrections without causing a mis-compare.
• Self-checked, shared-nothing transparent takeover
• Software fault tolerance• Message-based operating system
• Process pairs
• Transaction support
• Distributed single-system image
• Fault-tolerant parallel database
• Application server transaction processing monitors
= Logical synchronization units*
4-processor system
ServerNet technology
1203
1 20 3
1203
LSU*
120
3
Upgradeto triplex
SLIDE 619 July 2006
Integrity NonStop server delivers on the promise of outstanding value in its class
Infrastructure for thereal-time enterprise
• Addresses most demanding transaction processing requirements
• Exploits industry standards− Itanium processor, Java, SQL, HP
StorageWorks, HP OpenView
− HP Systems Insight Manager
• Simple, flexible management− Single application image
− OpenView
− Virtualized resources
• Unique Innovation for continuous application availability
− Integrated stack
− Linear scalability up to 4,080 processors
− Industry-leading data integrity
5 10
IBM
Sun $5,839
$7,185
* All figures in thousands of U.S. dollars.
Driving down IT operating costs with infrastructure innovations
Source: Standish Group, 2005
Lowest TCO*
$0 $500 $1,000
HP IntegrityNonStop
DMR
HP IntegrityNonStop
TMR
IBM zSeries(mainframe)
Basic cost Application cost
SLIDE 719 July 2006
NonStop Advanced Architecture (NSAA)
Application Virtualization across nodesApplication Virtualization across nodes
Software fault tolerance + DMR/TMR
Software fault tolerance + DMR/TMR
2–8 CPU/system2–16 CPU/node
NonStop hardware infrastructure
Modular I/O; limited connectivity No NonStop S-series I/O
(32 TB/node + XP + limited legacy)
NonStop hardware infrastructure
Modular I/O; virtually unlimited connectivity plus NonStop S-series I/O
(473 TB/node + XP + legacy devices)
4 or 8 GB main memory/CPU4, 8, or 16 GB main memory/CPU
~75% of NS16000 CPU (1.5 GHz/4 MB)
Highest CPU power (1.6 GHz/6 MB cache)
Scalable up to 2,040 processors(ServerNet internode)
Fully scalable up to 4,080 processors (P-Switch in node + ServerNet internode); lowest latency
Availability up to seven 9s Availability up to seven 9s
NS14000 server NS16000 server
Features at a glance
SLIDE 819 July 2006
Application Characterizations
• Integrity NonStop NS16000
− Applications requiring significant central processing capability > 8
processors
− Examples: Sabre OnLine Reservation, KDDI cellular SMS, large
bank ATM switch with ACI Base-24, large hospital (>400 beds)
supported by GE Centricity (nee IDX Carecast)
• Integrity NonStop NS14000
− High availability applications with limited scalability
− Examples: medium sized bank with Base-24, medium sized hospital
(150 – 400 beds) for GE Centricity (nee IDX Carecast)
SLIDE 1019 July 2006
Announcing the Integrity NonStop NS1000
Industry standard hardware components
NonStop software advantages
NonStop Value Architecture (NSVA)
=
+
Making NonStop more affordable
SLIDE 1119 July 2006
More choices:A portfolio of service levels
Ser
vice
leve
ls
Cost
Integrity NonStop 16000
Integrity NonStop 14000
NonStop Advanced Architecture
Integrity NonStop 1000
NonStop Value Architecture
High AvailabilitySystems
SLIDE 1219 July 2006
Itanium® NonStop and the NonStop Value Architecture (NSVA)
= Logical synchronization units*
NS16000/14000 4-processor system
ServerNet technology
1
2
0
3
1 20 3
1
2
0
3
LSU*
1
2
0
3
Upgradeto triplex
ServerNettechnology
NS1000 4-processor system
SNet PCI
Adapter
SNet PCI
Adapter
SNet PCI
Adapter
SNet PCI
Adapter
SLIDE 1319 July 2006
Itanium® NonStop and the NonStop Value Architecture (NSVA)
• Itanium2 Servers in a cluster of uni-processors
• 99.999% availability • NSVA built on NonStop principles
− Fault tolerant immediate "fail-fast" design paradigm
− Error detection on all data paths, memory, and disk
− Multiple paths to all data and communications
− Hot-pluggable components• Online repair • Online addition/upgrade
• Software fault tolerance• Message-based operating system• Process pairs• Transaction support• Distributed single-
system image
• Fault-tolerant parallel database
• Application server transaction processing monitors
ServerNettechnology
NS1000 4-processor system
SNet PCI
Adapter
SNet PCI
Adapter
SNet PCI
Adapter
SNet PCI
Adapter
SLIDE 1419 July 2006
HP Integrity NonStop NS1000
• HP rx2620 off the shelf
− Intel Itanium 2 processor at 1.3 GHz with 3 MB cache− Supports 2, 4, 6, 8 processors
− 4-8 GB memory per processor
• Software fault tolerance/fault isolation: same as other Integrity NonStop servers
− NonStop OS; NonStop data base SQL/MX
− Cluster programming transparency
− Software I/O checksum
• Industry standard Fibre Channel and Ethernet connectivity− Supports Storage Area Network (SAN) & Internal Storage
− Up to 32 TB storage per node (internal FC disks)
Most cost-effective, reliable, and easily managed software FT system
Delivered May 2006
NS100842
41 FCDM403938373635
34 NS I/O Core
33
32
31
30
2928
272625
2423
22
21**Console**
20
19 FCDM1817
1615141312
1110987
654
3
Configurable Space **
2
1 Maintenance Lan
Configurable Space
RX2620
RX2620
RX2620
RX2620
RX2620
RX2620
RX2620
RX2620
SLIDE 1519 July 2006
•Customers who have−High volume OLTP−High availability requirements−Medium to very high rates of data change
•Industry specific solutions−Small/med banking networks for ATM and point of sale (POS)
−Hospitals with fewer than 150 beds (GE Centricity)
−New application deployments in emerging countries
The NS1000: Brings significant TCO advantages to…
SLIDE 1619 July 2006
NS1000 Configuration Options
• 2p, 4p, 6p, 8p Hardware base
configurations (bundles)
• User can select additional
Hardware from the price list
e.g. data disks, etc.
• Upgrade Hardware bundles are
also available in 2p increments
• Software is not bundled and is
ordered from the NS1000 Software
price list
2 CPU 4 CPU 6 CPU 8 CPU
42
41 Configurable Space ** Configurable Space ** FCDM FCDM
40
39
38 FCDM FCDM FCDM FCDM
37
36
35 FCDM FCDM FCDM FCDM
34
3332
31
30
29
28 IOAM IOAM IOAM IOAM
27
26
25
24
23
22 Configurable Space Configurable Space Configurable Space Configurable Space
21 **Console** **Console** **Console** **Console**
20 rx2620
19
18 rx2620
17
16 rx2620 rx2620 rx2620 rx2620
15
14 rx2620 rx2620 rx2620 rx2620
13
12 rx2620 rx2620 rx2620
11
10 rx2620 rx2620 rx2620
9
8 rx2620 rx2620
7 Configurable Space **
6 rx2620 rx2620
5 Configurable Space **
4
3 FCDM FCDM
2
1 Maintenance Lan Maintenance Lan Maintenance Lan Maintenance Lan
SLIDE 1719 July 2006
NS1000 – a winning cost model
Source: Standish Group, 2006
Identical ATM application workload running on Wintel, Sun and IBM;
NS1000 cost data provided to Standish group for primary research.
$0
$100
$200
$300
$400
$500
$600
$700
$800
$900
$1,000
Annual TCO
($,000)
IBM AIX Sun Solaris Wintel NS1000
Systems
Application Cost
Basic Cost
SLIDE 1819 July 2006
NonStop Advanced Architecture (NSAA)
NonStop Value Architecture (NSVA)
Application Virtualization across systemsApplication Virtualization across nodes
Application Virtualization across nodes
Software fault tolerance (N+1 CPU FT)Software fault tolerance + DMR/TMRSoftware fault tolerance + DMR/TMR
2–8 CPU/system 2–8 CPU/system2–16 CPU/node
NonStop hardware infrastructure
Modular I/O; limited connectivity No NonStop S-series I/O
(32 TB/node + XP + limited legacy)
NonStop hardware infrastructure
Modular I/O; limited connectivityNo NonStop S-series I/O
(32 TB/node + XP + limited legacy)
NonStop hardware infrastructure
Modular I/O; virtually unlimited connectivity plus NonStop S-series I/O
(473 TB/node + XP + legacy devices)
4 or 8 GB main memory/CPU4 or 8 GB main memory/CPU4, 8, or 16 GB main memory/CPU
~65% of NS16000 server (1.3 GHz/3 MB)
~75% of NS16000 CPU (1.5 GHz/4 MB)
Highest CPU power (1.6 GHz/6 MB cache)
Scalable up to 8 processors with ServerNet(Expandable over Ethernet/ATM)
Scalable up to 2,040 processors (ServerNet internode)
Fully scalable up to 4,080 processors (P-Switch in node + ServerNet internode); lowest latency
Availability up to five 9s Availability up to seven 9s Availability up to seven 9s
NS1000NS14000NS16000
HP Integrity NonStop server family
SLIDE 1919 July 2006
NonStop Systems Performance Summary:Order Entry Benchmark
NonStop Systems Performance
Summary
0 1 2 3
NS16000
NS14000
NS1000
S88000
S86000
Series1
SLIDE 2119 July 2006
Comparison against Fundamentals:NonStop Server Products
S-Series NSVA
88K DMR TMR NS1000
Redundant input power feeds Y Y Y Y
Software fault tolerance Y Y Y Y
No single point of failure for a system Y Y Y Y
No single point of failure for a logical processor N N N N
Data integrity
End to end disk checksum Y Y Y Y
Transactional consistency for clean failover Y Y Y Y
Guaranteed Hardware Data Integrity Y Y Y N
CPU Fault masking - application transparent
Transient faults Good Better Best Good
Hard faults N Yes Yes! N
Service error resilient Good Better Best Good
SAN Fault Masking Y Y Y Y
NSAA
SLIDE 2219 July 2006
Impact of CPU hardware faults
No impactNo impactLose logical CPU
Lose logical CPUHard fault
Sometimes lose logical
CPU
Sometimes lose logical
CPU
No impact
NSVA
No impactNo impactNo impactTransient fault
(particle-induced)—main memory
No impactNo impactLose logical CPU
Transient fault (particle-induced)—
other parts of the CPU
No impactNo impactSometimes lose logical
CPU
Intermittent fault (underlying hardware defect)
TMRDMRNonStop S-series servers
Note: Logical CPU failure invokes software takeover, system keeps running.
SLIDE 2319 July 2006
Cascading Faults
• Most outages result from cascading faults rather than single faults−Example: a CPU halts (for any reason) and a defect in
recovery code in another CPU induces another halt
• Therefore, masking of single halts is beneficial even when the application is fault tolerant−Reduces the opportunities for having an outage
• DMR and TMR mask single hardware failures
• S series and NSVA do not
SLIDE 2419 July 2006
Impact of Repairing a CPU
• S series:−Loss of the logical processor
• NSVA:−Loss of the logical processor
• DMR:−No loss of logical processor(s)− Loss of fault tolerance and guaranteed data integrity for
all logical processors in that blade complex
• TMR:−No loss of logical processor(s)−No loss of fault tolerance for any logical processors in
that blade complex
SLIDE 2519 July 2006
20152009 2010 2011 2012 2013
HP NonStop Server Roadmap
2008200720062005 2014
S88000, S7800, S78
NS16000
Hebe
Perseus
Hera
NS16000CG
NS14000
Moirae
Electryon
NS1000
Triton
(Integrity NS Server)
(Integrity NS NEBs Server)
(Multi-core)
(Multi-core next generation)
(Blades implementation)
(S-Series Servers)
(Integrity mid-range NSAA Server)
(Mid-range refresh)
(Blades implementation)
(Integrity entry-level Server)
(Entry-level refresh)
NS5000 (-48V and CG servers)
NonStopAdvancedArchitecture
NonStopValue
Architecture