IBM eServer pSeries

IBM eServer pSeries
server
HCR
U6
server
pSeries
Revised April 2003
September, 2005
© 2005 IBM Corporation
IBM eServer pSeries
Ubicación del p615 en el cosmos de
servidores
© 2005 IBM Corporation
IBM eServer pSeries
Servidores de IBM
zSeries
The most reliable, mission-critical systems on
earth
pSeries
Most powerful, technologically advanced
UNIX servers
iSeries
High-performance integrated business servers
for mid-market companies
xSeries
Intel-based servers with mainframe-inspired
reliability technologies
© 2005 IBM Corporation
IBM eServer pSeries
El mercado de servidores UNIX
WW UNIX Server Revenue Share – Rolling 4Q
Average
40%
45%
35%
40%
30%
25%
35%
20%
IBM
HP
Sun
Other
30%
15%
10%
25%
5%
20%
15%
20
5
Q
10
5
Q
40
4
Q
30
4
Q
20
4
Q
10
4
Q
40
3
Q
30
3
Q
20
3
Q
10
3
Q
40
2
Q
30
2
Q
20
2
Q
5%
10
2
10%
Q
Quarterly UNIX-based Revenue Share
45%
Source: IDC Quarterly Server Tracker, 8/05
© 2005 IBM Corporation
IBM eServer pSeries
POWER4+: El motor del p615
© 2005 IBM Corporation
IBM eServer pSeries
pSeries 615 POWER4+ Chip Technology
•Superescalar.
Habilidad para ejecutar hasta
8 instrucciones en paralelo.
•Superpipelined.
Posibilidad de completar cinco
instrucciones por ciclo.
POWER4+ microprocessor
D
I
>1GHz CPU
>1GHz CPU
D
I
Shared L2 Cache
•L1 instruction cache
•128K/chip;64K/processor.
•L1 data cache
•64K/chip; 32K/processor.
•L2 cache
•1440K/chip; shared between
processors)
L3 Controller
L3 Directory
L3 Cache
Fabric Controller
Distributed Switch
Processor local bus
I/O Bus
Memory
.
© 2005 IBM Corporation
IBM eServer pSeries
pSeries 615 POWER4+ Chip Technology
Each processor contains
•two FXUs (fixed point units)
•two FPUs (floating point units)
•two load/store units
•a branching unit
•a CR (condition register) unit
© 2005 IBM Corporation
IBM eServer pSeries
POWER™ Microprocessor Roadmap
POWER4™
0.09 microns
0.13 microns
0.18 microns
1.0 to
1.3 GHz
Core
POWER5™
POWER4+™
1.2 to
1.9 GHz
Core
1.2 to
1.9 GHz
Core
>>> GHz >>> GHz
Core
Core
0.13 microns
1-1.3 GHz
Core
>> GHz
Core
1.4-2.0 GHz
CoreShared L2
Shared L2
Distributed Switch
Shared L2
Distributed Switch
2002-3
Distributed Switch
2001
Distributed Switch
Shared L2
LPAR
Autonomic computing
Chip multiprocessing
Shared L2
2005
Distributed Switch
2004
Larger L2
More LPARs
High-speed Switch
64-way SMP
Sub-processor partitioning
Enhanced Distributed Switch
Enhanced core parallelism
Improved floating-point performance
Faster memory environment
Dynamic firmware updates
© 2005 IBM Corporation
IBM eServer pSeries
POWER : The Most Scaleable Architecture
POWER2
PPC
401
PPC
750
PPC
750CXe
PPC
405GP
PPC
750FX
PPC
440GP
s
Server
PPC
750GX
PPC
970FX
PPC
440GX
op
t
k
s
e
D
s
Game
ed
d
d
e
b
m
E
Binary Compatibility
PPC
603e
POWER3
POWER4
POWER5
POWER4+
© 2005 IBM Corporation
IBM eServer pSeries
Anatomía del p615
© 2005 IBM Corporation
IBM eServer pSeries
Internal Structure of p615 Rack Model
Service
processor
Hot-plug power supplies
(one redundant – optional)
Six hot-plug PCI-X slots:
Three long 64-bit 133 MHz 3.3v
One short 64-bit 133 MHz 3.3v
Two short 32-bit 66 MHz 3.3v
3 hot-plug fans
(redundancy)
H C R U6
server
Two 4-packs
(separate backplanes)
hot-swappable
Ultra320 SCSI disks
1st four-pack standard
Server planar board contains:
pSeries
Op-panel
Three media bays
Two slimline
One standard
RJ-48 connector (when
in use, pre-empts the S1
serial port on rear)
One or two 2 1.2 GHz or 2-way
1.45 GHz POWER4+ processors,
1.5MB L2 Cache, 8MB L3 Cache,
1 to 16GB DDR Chipkill memory,
Three serial ports, two optional
HMC ports, keyboard, mouse and
parallel ports
10/100 and 10/100/1000 (1 Gbps)
Ethernet ports
© 2005 IBM Corporation
IBM eServer pSeries
Rear View of p615 Rack Model
server
H C R U6
server
Two hot-plug AC power
supplies with dual
isolated line cords
pSeries
Three serial ports (S1, S2, S3)
Mfg. Diagnostic port (not for client use)
Parallel
Rack
indicator
light port
S2
S3
S1
HMC 1
HMC 2
10/100 Ethernet
10/100/1000 Ethernet
Six hot-plug PCI-X slots:
Three long 64-bit 133 MHz 3.3v
One short 64-bit 133 MHz 3.3v
Two short 32-bit 66 MHz 3.3v
© 2005 IBM Corporation
IBM eServer pSeries
p615 Memory Quad Positioning
Main memory ranges from 1GB to 16GB ECC DDR SDRAM.
Memory is configured in two quads on the planar board,
and is pre-balanced by design for optimal performance with
each quad split between two synchronous memory interface (SMI) controllers.
Plan ahead to reach desired maximum memory to avoid discarding memory.
1GB quad = 4 X 256MB DIMMs
2GB quad = 4 X 512MB DIMMs
4GB quad = 4 X 1GB DIMMs
8GB quad = 4 X 2GB DIMMs
SMI
L3
Shared L2
Distributed switch
Quad 1
Quad 2
SMI
© 2005 IBM Corporation
IBM eServer pSeries
pSeries 615 Server Peak Bandwidth per
1.2 GHz 2-way Chip
12.8GB/sec (= 2 x 16 bytes @ 1/3 Clock Speed)
For 2-way chip
L3
Shared L2
6.4GB/sec
(= four 8-byte
paths @ 200 MHz)
M
E
M
O
R
Y
1 to 16GB
GX Bus
3.2GB/sec (= dual 4-byte paths @ 1/3 Clock Speed)
Distributed switch
Six PCI-X slots
I/O Hub
2GB/sec
© 2005 IBM Corporation
IBM eServer pSeries
pSeries 615 Server Peak Bandwidth per
1.45 GHz 2-way Chip
15.5GB/sec
For 2-way chip
L3
6.4GB/sec
Shared L2
M
E
M
O
R
Y
1 to 16GB
GX Bus
3.9GB/sec
Distributed switch
Six PCI-X slots
I/O Hub
2GB/sec
© 2005 IBM Corporation
IBM eServer pSeries
RAS
(Fiabilidad, Disponibilidad y Servicialidad)
© 2005 IBM Corporation
IBM eServer pSeries
POWER4+ Reliability
Across the Product Line
24x7
IBM
pSeries
server
Mainframe-inspired RAS
Reliability,
Availability and
Serviceability
SelfConfiguring
SelfHealing
SelfSelfOptimizing Protecting
ris
e
S
p
ris
e
S
p
H C R U6
server
server
H C R U6
server
pSeries
First Failure Data Capture
DDR ECC Chipkill™ memory
Bit-steering/redundant memory
Memory soft scrubbing
Redundant power, fans
Dynamic Processor Deallocation
Deallocate PCI-X bus, L2/L3 cache
Persistent memory deallocation
Hot-plug PCI-X slots, fans, power
Internal LED diagnostics
Hot-swappable disk drives
© 2005 IBM Corporation
IBM eServer pSeries
First Failure Data Capture Technology
Old Failure Recreate Strategy
–Run diagnostic test cases during service call
Fault Isolation Register (FIR) –Repair based on test case symptom
(unique fingerprint of each –Questionable correlation to original problem
error captured)
–Development focus on improved test cases
–Open service action plan if failure not recreated
–Test cases used to try to isolate failures and verify
Service
correct operation
Processor
Error Checkers
CPU
L1 Cache
L2/L3 Cache
Log Error
Memory
Nonvolatile
RAM
Disk
Reliably identify each failing component,
reducing costly downtime
First Failure Data Capture: an IBM Exclusive
–Specialized hardware designed to capture failure
data at time of failure
–Repair based on root-cause analysis
–Direct correlation to original client problem
–Engineering focus on built-in error detection and
capture
–Service action plan driven by captured failure
information
–Test cases used to only to verify correct operation
© 2005 IBM Corporation
IBM eServer pSeries
Start Off With Quality-based Design
lt
Fau nce
ida
Avo
Designing extra quality into
system to keep errors from
ever happening
Mainframe-class components and technologies
- Reduced power consumption, cooler operating
temperatures for increased reliability
© 2005 IBM Corporation
IBM eServer pSeries
ECC Memory Integrity and Availability
Memory scrubbing for soft single-bit errors
that are corrected in the background while
memory is idle, to help prevent multiple-bit
errors.
Failing memory bit steered
to spare memory chip
Chipkill
X
X
Dynamically reassign memory I/O
via bit-steering if error threshold
is reached on same bit
....
Spare
memory
chip
Scatter memory chip bits across four separate ECC words for Chipkill recovery
Bit-scattering allows normal single bit ECC error processing, thereby
keeping system running with a Chipkill failure.
Bit-steering allows memory lines from a spare memory chip to be
dynamically reassigned to a memory module with a faulty line to
keep system running.
If all bits are used up on the spare memory chips, and threshold is reached,
the Service Processor will be invoked to request deferred maintenance at a time
acceptable to client.
© 2005 IBM Corporation
IBM eServer pSeries
Built-in Redundancies for Non-stop Operation
N+1 ancy
und
d
e
R
Use of redundancy to remain
operational with full resources
Redundant spare memory chips
Redundant fans
- Fans will speed up to try to compensate for failed fan
Redundant power supplies (optional)
© 2005 IBM Corporation
IBM eServer pSeries
p615 PCI Bus and Card Error Recoveries
PCI-X
Controller
X
Problem with damaged connection results in PCI-X bus error
- That particular slot is varied offline by system
- All other slots remain active
- System stays operational without the slot active
- If PCI card is damaged, replace with new one
- If slot is damaged, plan for maintenance at client’s convenience
© 2005 IBM Corporation
IBM eServer pSeries
Replacement of Parts While System Runs
t
ren
r
u
c
e
Con enanc
nt
M ai
Online hot-plug of defective
electromechanical components
keeps system operational
Disk drives
Cooling fans
Power subsystems
PCI, PCI-X adapter cards
© 2005 IBM Corporation
IBM eServer pSeries
Quicker Service for Client Convenience
vice
Se r
y
tivit
c
u
d
Pr o
Increasing service productivity means
more client uptime
Internal LED diagnostics identify components that
require service
LEDs on I/O provide status of PCI-X slots and
disk drives for power, hot-swap and need for service
© 2005 IBM Corporation
IBM eServer pSeries
Sistemas Operativos
© 2005 IBM Corporation
IBM eServer pSeries
AIX 5L™ & Linux
AIX
Linux Servers
AIX
Hypervisor
Linux
server
HCR
U6
server
pSeries
pSeries
server
Native 64-bit Linux
High performance POWER4
SMP, LPAR, Clusters
Proven, reliable pSeries
Surprisingly affordable
compared to Intel SMPs
Standard Linux APIs in AIX
Common Functionality
Interoperability Testing
AIX Toolbox for Linux Apps
- Linux development
environment on AIX
Run Linux and AIX on the same
pSeries server
Workload Consolidation - Linux
for Infrastructure, AIX for
Database
Server Consolidation - Develop,
Test, Production
Competitive Consolidation
© 2005 IBM Corporation
IBM eServer pSeries
Referencias
© 2005 IBM Corporation
IBM eServer pSeries
For more details, see the POWER4
System Microarchitecture white
paper at http://www1.ibm.com/servers/eserver/pseries/h
ardware/whitepapers/power4.html
(or search for "power4 architecture"
in the Search window at
www.ibm.com)
–
–
–
© 2005 IBM Corporation
IBM eServer pSeries
Y ya está...
(a no ser que tengais más preguntas)
© 2005 IBM Corporation