Five Selling Points of Huawei Servers - ActForNet · 2014-11-18 · Note: Huawei servers such as...
Transcript of Five Selling Points of Huawei Servers - ActForNet · 2014-11-18 · Note: Huawei servers such as...
Five Selling Points
of Huawei Servers
1
Where Are Servers Sold?
Server classification
Large data centers of public service providers
Internal data centers of customers from government,
finance, energy, and manufacturing industries
Equipment rooms of small- and medium-sized
enterprises (SMEs) and research institutes
★ As computing is required everywhere, servers are required
everywhere. Servers are necessary for IT construction.
★ Wherever there are storage and network, there are servers.
No matter how small its volume is, data must be processed by servers
before it is transmitted and stored.
High-
density
servers
Rack
servers
Blade
servers
2
What We Sell of Servers?——Performance,
Reliability, Huawei Proprietary and Controllability
Performance
Virtualization consolidation
Data processing
Mission-critical applications
Stability without downtime
• Huawei proprietary design
• Huawei proprietary core
chips
Reliability
Huawei
proprietary and
controllability
Focus on strategic NA, closely cooperate with ISV and DP to break through
core-value industries
3
Five Selling Points of Huawei Servers
Low failure rate: More than 500,000 servers are delivered and used in a large scale with
the failure rate 15% lower than the average industrial level.
Outstanding performance: Huawei servers have shattered 83 world records for
computing performance and energy efficiency.
High security and controllability: Huawei is the only vendor that supports one-stop
R&D, production, and delivery and uses Huawei proprietary core chips.
High availability: With rich RAS features, Huawei servers can stably run in the
temperature of 40°C for a long term.
Continuous innovation: Huawei has strong R&D capability and uses Huawei proprietary
core chips for accelerating applications and improving service performance.
4
1. Failure Rate 15% Lower than the Average Industrial Level
• Stable connection to services and reliable system
running are the most concerned issues for CIOs.
• Failure rates of servers and components in the
servers must be controlled in an acceptable range.
Passive backplane
The components of the
backplane include only
a PCB and connectors
rather than any active
component such as a
resistor or IC. This
prevents single point of
failures (SPOFs).
Higher standards for the
reliability test
Reliability test with standards higher
than industrial standards
High-availability
components
Carrier-class components
Life time over 5 years
Long-term stable running
Strict quality control
Strict quality control through
component purchase, board R&D,
and system test
1. The low failure rate reduces service interruption
and data loss caused by system breakdowns and
component failures.
2. The low failure rate lowers the system O&M cost.
Tested at the Tencent data center, Huawei servers
have the lowest defective parts per million (DPPM)
value.
Customer Case
Customer Benefits
5
2. Outstanding performance: 83 World Records for Computing Performance and Energy Efficiency
• Based on the same x86 platform, CPUs, and memory, different
servers provide different performance.
• Running the same services, different servers consume
different electric power.
Root cause of differences: hardware and software
optimization
Leading computing
performance
Huawei V2 servers
shattered 74 SPEC
records on integer and
floating-point computing
performance.
Energy saving
All-round energy-saving technologies
to ensure optimal energy efficiency
Database No. 1 TPC-E
performance
First Chinese TPC member
Hardware and software
optimization
High power conversion efficiency (AC-
DC and DC-DC)
BIOS computing performance
optimization and dynamic adjustment
1. With the same platform and configuration,
Huawei servers provides stronger computing
capability than competitors' products.
2. Huawei servers use all-round energy-saving
technologies to reduce electricity costs.
1. Huawei servers won highest scores in technical
tests in centralized procurement of ICBC, China
Mobile, and State Grid.
Customer Case
Customer Benefits
6
3. High Security and Controllability: Unique Vendor That Supports One-Stop R&D, Production, and Delivery
Huawei proprietary chips
(management chips and NC chips) Hardware schematic diagram
PCB interconnection
(backplane switching)
Heat dissipation simulation
BIOS/BMC and CPLD/FPGA iMana (management software)
Basically, competitors in China do not have comprehensive R&D systems. Most of them provide servers
manufactured by ODM vendors in Taiwan and United States.
Competitors such as HP, IBM, Dell, and Cisco design architecture and management software and have ODM
vendors such as Foxconn, Inventec, Quanta for the design, test, and production.
1. Huawei uses self-developed products to
provide customized, optimized servers.
2. Huawei servers are not OEM products,
which ensures end-to-end quality and
delivery.
0
Sugon servers are OEM products and tested
with defects in the China Mobile's centralized
procurement test. The rectification requires a
long time because Sugon products rely on
Taiwan ODM vendors.
Customer Case
Customer Benefits
7
Competitors' servers operate in the range of 10°C to 35°C. Huawei mainstream
servers operate in the range of 5°C to 40°C.
4. High Availability: High Adaptability and Stable Running in 40°C
Competitors' server operating
temperature range 10℃ to 35℃ Huawei RH2285 operating
temperature range 5℃ to 40℃
The 40°C specifications are higher
than that of all competitors except
IBM. Huawei servers use the same fans
as those of IBM servers. However, IBM
servers support the 40°C specifications
under specific conditions.
In the temperature range of 5°C to 35°C, the Huawei RH2288H
V2 supports CPUs (80 W to 130 W) to work at the normal
frequency (without frequency decrease) when one fan fails;
however, when one fan fails or in the temperature higher than
30°C, CPU performance of HP servers is affected.
The Dell R720 and R720xd support the one-fan failure under
specific conditions.
Note: Huawei servers such as the RH2285 V2, RH2288 V2, RH2485 V2, E9000, XH310 V2, and XH311 V2 support
the high temperature of 40°C. Other Huawei servers support this temperature under specific conditions.
1. Huawei servers can operate in a
maximum temperature of 40°C, which
reduces cooling costs of the equipment
room.
2. Huawei servers provide high availability
to ensure service continuity for
customers.
1. When air conditioners in the Ministry of
Railways (MOR) equipment room fail,
competitors' servers break down as the
temperature in the equipment room rises.
However, Huawei servers are running
normally.
Customer Case
Customer Benefits
8
4. High Availability and Rich RAS Features
Dynamic adding or removing
nodes • When a 4-socket secondary node
is faulty, you can replace the faulty
parts or the whole faulty node to
rectify the fault.
• Node resources can be added or
removed dynamically based on
service load requirements to
optimize configuration.
Two-dimension hot swap • You can replace PCIe cards
without opening the chassis and
powering off the server.
• The replacement time is
shortened from 10 minutes to 1
minute, increasing efficiency.
Fault tolerance for integrated
hardware and software • System-grade fault tolerance for the
integrated design of management
software, BIOS, hardware, and chips
• 35 RAS features
• Fault-tolerant architecture with fault
detecting, diagnosis, and isolation
DIMM fault precaution and
hot swap • Fault precaution and precise fault
location facilitate replacement of
faulty DIMMs (DIMM hot swap).
• The server system support Mirror or
Spare mode.
• Online memory expansion meets
requirements for service load
increase.
Hot removal for nodes
Hot adding
for nodes
Fault
management
architecture
ereport
error fault fault
Hardware fault
injection
Software
simulation
injection
Virtual error
event injection
Fault
alarm fault
Core isolation
Page isolation
DIMM hot swap
I/O device hot swap
Application system
Linux Kernel
Hardware
Fault statistics and analysis
Faulty device location
Cause analysis
Expert system Fault response
Error log
Fault log User interface
Kernel matches with the FMD interface,
unified error transmission channel
Hardware fault information capture and
fault handling
CPUs DIMMs PCIe
devices xxx
Fault i
nje
ction
9
5. Continuous Innovation: Huawei Proprietary Core Chips to Accelerate Applications and Improve Service Performance
• Service bottlenecks focus on I/O. Computing
performance is sufficient. However, I/O
bandwidth is insufficient.
• Limited hard disk read and write bandwidth and
disunified ports
CPU interconnection chips
The Node Controller 1.0 is approved
and accepted in China. The new-
generation NC 2.0 (interconnection x86
CPU) is being developed.
Server management
chip
Single chip, low power
consumption, and
wireless management
features
Huawei proprietary I/O chips
Converged 10GE/RoCE/FCoE/FC
RAID controller
I/O acceleration card
PCIe SSD card with industry-leading
performance
iNIC with excellent virtualization
performance
1. The 8-socket or higher system provides higher
performance and reliability.
2. The I/O acceleration solution provides higher
bandwidth to fully realize the CPU potential.
The Deltares uses the E9000. The customer said that the
HP C7000 is too plastic compared with the Huawei E9000.
The 32-socket fault-tolerant computer that uses the NC 1.0
chips is approved and accepted by the Ministry of Science.
The Huawei E9000 with the iNIC provides the highest
SPECvirt virtual performance.
Huawei SSD cards got the first place in the StorageReview
test.
Customer Case
Customer Benefits
10
Supportive Materials for Five Selling Points
11
1. Failure Rate 15% Lower than the Average Industrial Level
• Stable service connection and reliable system running are the
most concerned issues for CIOs.
• Failure rates of servers and components in the servers must
be controlled in an acceptable range.
Passive backplane
The components of the
backplane include only a
PCB and connectors rather
than any active component
such as a resistor or IC. This
prevents single point of
failures (SPOFs).
Higher standards for the
reliability test
Reliability test with standards
higher than industrial standards
High-availability
components
Carrier-grade components
Life time over 5 years
Long-term stable running
Strict quality control
Strict quality control through
component purchase, board R&D,
and system test
1. The low failure rate reduces service
interruption and data loss caused by system
breakdowns and component failures.
2. The low failure rate lowers the system O&M
cost.
Tested at the Tencent data center, Huawei servers
have the lowest DPPM value.
Customer Case
Customer Benefits
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
12
Supportive Point 1: Passive Backplane with 0 Failure Rate
Huawei's passive backplanes Competitors' active backplanes
Huawei's passive backplanes are operational without power
supply. The passive backplanes contain no active components
such as resistors, capacitors, and ICs, except PCBs and
connectors. They are free from SPOFs.
Competitors' active backplanes are operational only with
power supply. The active backplanes contain active
components such as resistors, capacitors, and ICs, besides
PCBs and connectors. This design has the risks of SPOFs.
The passive backplanes effectively prevent operation exceptions in the chassis caused by faulty active components
and ensure high reliability thanks to its appropriate zero failure rate.
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
13
Supportive Point 2: Reliable Components to Support Stable Running in Harsh Environments
Carrier-class components and reliability design
Huawei SMD macromolecular capacitors Competitor electrolyte plug-in capacitors
Carrier-class derating design to ensure stable running
in the high temperature of 40°C
Components with a life time of over 5 years to
support long-term, reliable running
Macromolecular components instead of electrolyte to
ensure safety
Anti-interference capability and surge protection
better than competitors' products
EMC Competitiveness Comparison
Huawei Competitor
Conducted susceptibility 10 V 3 V
Signal EFT 1 KV 0.5 KV
Power supply EFT 2 KV 1 KV
Radiated susceptibility 10 V/M 3 V/M
ESD susceptibility Contact discharge 6 KV/Air
discharge 8 KV
Contact discharge 4 KV/Air
discharge 8 KV
Anti-surge capability ITU 2.5/2.5 KV IEC 2 KV/2 KV
Radiation emission (RE) CE/FCC CLASS A CE/FCC CLASS A
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
High anti-interference capability
14
Power-on
Power-off
Time
48 h 120 h 48 h
Supportive Point 3: Strict Standards for the Reliability Tests to Minimize the Failure Rate
Reliability tests are complete between
TR4A and TR5 (one month required).
Totally three cycles of reliability tests
are complete.
A 48-hour power cycle test
and a 120-hour pressure
test are complete within
one cycle.
Stage 1 reliability tests Stage 2 reliability tests
TR4 TR4A TR5
Huawei servers are tested by the reliability tests with standards higher than industrial standards. The reliability test consists of
two stages. The first stage is TR4 to TR4A. The second stage is TR4A to TR5. In the first stage, test items are performed separately
on multiple servers or centrally on one server in the reliability test environment. In the second stage, test items are centrally performed
on one server in the reliability test environment.
Stage 1
reliability tests
Power cycle test (power on and off
once) in the normal temperature
Power cycle test (BMC controls
power-on/off) in the normal
temperature
Power-off and power-
on during POST in the
normal temperature
POST reboot test
in the normal
temperature
OS reboot test in the
normal temperature
Pressure test in the
normal temperature
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
15
Supportive Point 4: Strict Quality Control Throughout the Purchase, R&D, and Tests
Supplier certification, component selection, and quality control
EMC laboratory, vibration test laboratory, high temperature test
laboratory, and safety regulation laboratory
Huawei's fault-tolerant architecture provides comparable reliability
as RISC midrange computers.
Huawei has the design capability for architecture with 99.999%
reliability for high-end products.
Board design, server design, and the test process meet Integrated Product
Development (IPD) standards.
35 Reliability, Availability, and Serviceability (RAS) features are supported.
150 patents on reliability designs are obtained.
200,000 carrier-class server blades have been delivered to telecom
users.
300,000 rack servers, blade servers, and high-density servers have
been delivered to enterprise users.
Precise design
Strict tests
500,000
servers
in large-scale
use
Laboratories with sufficient equipment
Outstanding reliability in both components
and systems
High reliability
Architecture design capability
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
16
Supportive Point 5: Failure Rate Comparison Between Huawei and Competitor products at T Customer Site
2012 January February March April May June July August September October November December
Monthly failure rate of Huawei blade servers (without hard disks) 0.06% 0.07% 0.09% 0.04% 0.10% 0.10% 0.10% 0.08% 0.05% 0.07% 0.05% 0.06%
Monthly failure rate of servers from D (without hard disks) 0.11% 0.15% 0.05% 0.10% 0.13% 0.08% 0.13% 0.10% 0.11% 0.16% 0.07% 0.09%
Monthly failure rate of servers from L (without hard disks) 0.23% 0.31% 0.20% 0.34% 0.36% 0.25% 0.27% 0.42% 0.34% 0.29% 0.25% 0.33%
Monthly failure rate of Huawei blade servers
(without hard disks)
Monthly failure rate of servers from D
(without hard disks)
Monthly failure rate of servers from L
(without hard disks)
January February March April May June July August September October November December
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
17
2. Exllent performance: 83 World Records for Computing Performance and Energy Efficiency
• Based on the same x86 platform, CPUs, and memory,
different servers provide different performance.
• Running the same services, different servers consume
different electric power.
Root cause of differences: optimization of software and
hardware
Leading computing
performance
Huawei V2 servers shattered
74 SPEC records on integer
and floating-point computing
performance.
Energy saving
All-round energy-saving technologies
to ensure optimal energy efficiency
Database No. 1 TPC-E
performance
First Chinese TPC member
Hardware and software
optimization
High power conversion efficiency (AC-
DC and DC-DC)
BIOS computing performance
optimization and dynamic adjustment
1. With the same platform and configuration,
Huawei servers provides stronger computing
capability than competitors' products.
2. Huawei servers use all-round energy-saving
technologies to reduce electricity costs.
1. Huawei servers won highest scores in technical
tests in centralized procurement of ICBC, China
Mobile, and State Grid.
Customer Case
Customer Benefits
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
18
Supportive Point 1: 83 World Records for Computing Performance and Energy Efficiency
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
• High computing performance: 74 world records of integer and floating-point computing performance in the SPECint/fp test
• Energy-saving: 9 world records in the SPECpower test
Huawei servers have
shattered 83
world records in the SPEC
benchmark test by now.
19
Supportive Point 2: Unique Chinese Vendor That Offers TPC Test Capabilities
Transaction Processing
Performance Council (TPC)
2012. 01 2012. 12 2013. 02 2013~
RH5885 (1 TB memory)
No. 1 in the TPC-E test
• Huawei joined TPC-E/H/DS/DI.
• Huawei took part in defining the TPC-DI benchmark.
• Guiding new
standards
• Released results
for new products
First Chinese TPC Member
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
20
Supportive Point 3: RH5885 Ranked First in the TPC-E Test
1 TB memory
• 8 x E7-8800 CPUs
• 128 DIMM slots
• 16 x 2.5" hard disks
• 4 x E7-4800 CPUs
• 64 DIMM slots
• 8 or 10 x 2.5" hard disks
RH5885 V2
(4-socket)
RH5885 V2
(8-socket)
TPC-E test story
After the Huawei 4-socket RH5885 V2 got 3054 tpsE in the TPC-E test, IBM reconfigures its x3850 X5 to achieve the 3218 tpsE score.
With 1 TB memory, the Huawei RH5885 V2 achieves higher score in the TPC-E test than the IBM x3850 and HP DL580 G7.
3,218 3,054
2,827
2,455 2,651
0
500
1000
1500
2000
2500
3000
3500
IBM 4S x3850 X5 (2TB)
Huawei 4S RH5885 V2 (1TB)
IBM 4S x3850 X5(1TB)
HP 4S DL580 G7 (1TB)
Fujitsu 4S RX500 G7 (512GB)
TPC-E Performance (tpsE)
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
21
Supportive Point 4: All-round Energy-saving Technologies to Ensure Optimal Energy Efficiency
Huawei dynamic energy management
technology (DEMT) Energy-saving cooling design
1. Vector airflow management
2. Double-faced cellular boards
3. Aluminium board embedded
with copper heat sink
4. Unique air channel design
Highly efficient PSUs
Dynamic power capping
Energy-saving and performance-
optimized OS
Device hibernation and awakening
Energy
efficiency
No.1
1. CPU hibernation
2. Hard disk
hibernation
1. High-efficiency POL
module
2. 80 Plus PSUs
3. HVDC power supply
1. Dynamic disabling of CPU cores
2. CPU frequency and voltage
adjustment
3. Dynamic memory frequency
adjustment
1. Dynamic phase closing for power
supply
2. Non-simultaneous power-on hard
disks
3. Multiple-input multiple-output (MIMO)
4. Non-simultaneous power-on server
blades
3. PSU hibernation
4. System
hibernation
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
1. Energy reservation
menu in the BIOS
2. OS performance
optimization
22
Supportive Point 5: Huawei Server Energy-saving Technologies Recognized Worldwide
Huawei full series of rack servers passed the Energy Star, China Environmental Labelling, and Energy Conservation and Environmentally-friendly Certification released by China Quality Certification Center (CQC)certification. Huawei
is the first vendor that obtained the China Environmental Labelling certification.
Country/Region EU North America China International Russia Australia Nigeria Saudi
Arabia Mexico
Compulsory
Certification REACH-DOC ROHS-DOC CE-SDOC WEEE
UL
FCC
(United States)
IC
(Canada) CCC CB GOST C-TICK SONCAP SASO UL-COC NOM
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
23
3. High Security and Controllability: Unique Vendor That Supports One-Stop R&D, Production, and Delivery
Huawei proprietary chips
(management chips and NC chips) Hardware schematic diagram
PCB interconnection
(backplane switching)
Heat dissipation simulation
BIOS/BMC and CPLD/FPGA iMana (management software)
Basically, competitors in China do not have comprehensive R&D systems. Most of them provide OEM servers
from Taiwan and United States.
Competitors such as HP, IBM, Dell, and Cisco design architecture and management software and have ODM
vendors such as Foxconn, Inventec, Quanta for the design, test, and production.
1. Huawei uses self-developed products to
provide customized, optimized servers.
2. Huawei servers are not OEM products,
which ensures end-to-end quality and
delivery.
0
Sugon servers are OEM products and tested
with defects in the China Mobile's test. The
rectification requires a long time because Sugon
products rely on Taiwan ODM vendors.
Customer Case
Customer Benefits
Low failure rate
High availability Continuous innovation
Security and reliability
Outstanding energy
efficiency
24
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
Competitors' servers operate in the range of 10°C to 35°C. Huawei mainstream
servers operate in the range of 5°C to 40°C.
4. High Availability: High Adaptability and Stable Running in 40°C
Competitors' server operating
temperature range 10℃ to 35℃ Huawei RH2285 operating
temperature range 5℃ to 40℃
The 40°C specifications are higher
than that of all competitors except
IBM. Huawei servers use the same fans
as those of IBM servers. However, IBM
servers support the 40°C specifications
under specific conditions.
In the temperature range of 5°C to 35°C, the Huawei RH2288H
V2 supports CPUs (80 W to 130 W) to work at the normal
frequency (without frequency decrease) when one fan fails;
however, when one fan fails or in the temperature higher than
30°C, CPU performance of HP servers is affected.
The Dell R720 and R720xd support the one-fan failure under
specific conditions.
Note: Huawei servers such as the RH2285 V2, RH2288 V2, RH2485 V2, E9000, XH310 V2, and XH311 V2 support
the high temperature of 40°C. Other Huawei servers support this temperature under specific conditions.
1. Huawei servers can operate in a
maximum temperature of 40°C, which
reduces cooling costs of the equipment
room.
2. Huawei servers provide high availability
to ensure service continuity for
customers.
1. When air conditioners in the Ministry of
Railways (MOR) equipment room fail,
competitors' servers break down as the
temperature in the equipment room rises.
However, Huawei servers are running
normally.
Customer Case
Customer Benefits
25
Heat competitiveness comparison
RH2285H IBM 3650
Temperature 5°C to 40°C 10°C to 35°C
Dimensions (H x W x D) 86.1 mm x 447 mm x 740 mm 85.2 mm x 443.6 mm x 698 mm
Heat Dissipation
Capability 800 W 675 W
SPECpower 5088 4708
Key technologies:
• Vector airflow management: ensures heat dissipation for high-power CPUs (95 W/130
W/135 W) by efficiently using limited airflow.
• Double-faced cellular board: increases the amount of system airflow input by achieving
the open porosity of 43%, which increased 115% compared with HP's products.
• MIMO dynamic energy control: reduces energy consumption by 4.22%, raising the
ranking position from No.6 to No.1.
• Heat sink of aluminum substrate embedded with copper: reduces the heat sink weight
by 45%, cutting the cost by 40%.
Advantages:
• Huawei servers rank first in the industry SPECpower test.
• The heat dissipation capability of Huawei servers supports smooth evolution of servers, three-to-five-year
evolution of the architecture, and two-generation evolution of CPUs.
• Huawei servers can run at 40°C. The general running temperature in the industry is 35°C.
Multiple-input multiple-output
(MIMO) dynamic energy control
Heat sink of aluminum
substrate embedded
with copper
Aluminum substrate Copper
Double-faced cellular board
System vector
airflow management
CPU
Supportive Point 1: Vector Airflow Management, Dynamic
Energy Control, Full Series CPUs Running at 40°C
Low failure
rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
26
Supportive Point 2: The MIMO Technology
Saves Energy by 5%
Fan control
Temperature (chips,
Ta) CPU states
Fan speed
Fan speed
Target temperature
Noise limitation
Temperature limitation
CPU states
Input Output
Background:
When the traditional fan speed adjustment technology is used, the fan speed is minimized
when the device is operating at the normal temperature, ensuring lowest noise while
implementing heat dissipation.
As the x86 manufacturing process develops (especially for the process of less than 45 ns),
the leakage current of CPUs increases with the temperature. For example, when the
temperature of a CPU developed on the SNB platform increases by 1°C, the power
consumption of the CPU increases by about 0.7 W.
The fan power consumption is proportional to the fan speed by cube.
P_total = P_blade + P_fan
P_total: Specifies the device power consumption.
P_blade: Specifies the hardware power consumption.
P_fan: Specifies the fan power consumption.
The higher the fan speed, the lower the component temperature and the smaller the
leakage current, that is, the smaller P_blade.
The higher the fan speed, the higher the fan power consumption, that is, the higher
P_fan.
MIMO heat dissipation and energy saving technology:
The MIMO technology dynamically controls the fan speed in real time, balances the fan
power consumption and hardware power consumption, and optimizes the system power
consumption based on parameter value changes.
The input parameters for energy control include the temperature (temperature of chips and
environment), CPU running status, fan speed, and thresholds (noise/temperature). The
output parameters include the target temperature of components, fan speed, and CPU
status.
Energy saving effect:
The MIMO technology can save energy by 5%, which is higher than the traditional fan speed
adjustment technology.
Relationship between the CPU
leakage current and temperature
Fan speed
Syste
m p
ow
er
CPU high utilization
CPU low utilization
Relationship between the optimal
energy saving point and fan speed
Low failure
rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
27
Partition board between half-width slots
PSU air channel
Half-width slot
LCD guide rail area
Air channel for rear modules
Front server blade area
PSU air channel
Supportive Point 3: Independent Air Channels Increase Heat Dissipation Efficiency
Independent PSU
cool air channel
Three fan
modules
support N+1
redundancy.
Independent air channels for the PCIe card:
Z-shaped air channel and C-shaped air channel
Independent air channel
for server nodes
X6000 high-density server
Independent air channels
E9000 blade server
Low failure
rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
28
5. Continuous Innovation: Huawei Proprietary Core Chips to Accelerate Applications and Improve Service Performance
• Service bottlenecks focus on I/O. Computing
performance is sufficient. However, I/O
bandwidth is insufficient.
• Limited hard disk read and write bandwidth and
disunified ports
CPU interconnection chips
The Node Controller 1.0 is approved
and accepted in China. The new-
generation NC 2.0 (interconnection x86
CPU) is being developed.
Server management
chip
Single chip, low power
consumption, and
wireless management
features
Huawei proprietary I/O chips
Converged 10GE/RoCE/FCoE/FC
RAID controller
I/O acceleration card
PCIe SSD card with industry-leading
performance
iNIC with excellent virtualization
performance
1. The 8-socket or higher system provides higher
performance and reliability.
2. The I/O acceleration solution provides higher
bandwidth to fully realize the CPU potential.
The Deltares uses the E9000. The customer said that the
HP C7000 is too plastic compared with the Huawei E9000.
The 32-socket fault-tolerant computer that uses the NC 1.0
chips is approved and accepted by the Ministry of Science.
The Huawei E9000 with the iNIC provides the highest
SPECvirt virtual performance.
Huawei servers got the first place in the SSD
StorageReview test.
Customer Case
Customer Benefits
Low failure rate
Outstanding energy
efficiency
Security and reliability
High availability Continuous innovation
29
Supportive Point 1: Continuous Innovation, Huawei-Developed Key Server Chips
Out-of-band management chip (BMC)
• Adopts the single-chip solution to reduce power
consumption and space.
• Implements stateless computing and wireless
management.
Hadoop acceleration card
• Uses the hardware uninstallation GZIP
algorithm, decreasing the CPU usage by 20%
and compression and decompression power
consumption by 80%.
• Provides a standard ZLIB to enable upper-layer
services to use hardware compression cards
transparently.
Convergence I/O chip
• 10 GE, RDMA (10 GE), TOE, FCoE, FC
8G/16G all in one
• High cost-effectiveness
PCIe SSD controller chip
• SSD controller that provides the SAS/PCIe
interface.
• Supports PCIe 3.0 and SAS 3.0.
RAID controller chip
• Improves I/O performance and supports the
supercapacitor, HA backup, and PCIe SSD
controller.
• Implements 3D networking, secure policies, and
return on capital employed (RoCE) low-latency
network.
• Supports high-performance, low-cost virtualization
network solutions.
Future-oriented network chip
Low failure
rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
30
Supportive Point 2: Highly Efficient Interconnection Chips for Multiple Servers
Meets requirements for high-
performance OLTP database
applications and provides low-
latency, high-bandwidth
interconnecting architecture for 8-
socket to 64-socket servers that use
Xeon CPUs.
Supports migration of
interconnection of massive CPU
nodes and provides key
technologies based on scale out
and scale up convergence to
support low-latency interconnection
expansion.
NC
CPU
DIMM
CPU
NC
CPU
DIMM
CPU
NC
CPU
DIMM
CPU
Zone 3 Zone 4 Zone N
NC
Zone 1
NC DIMM DIMM
CPU CPU
NC
Zone 2
NC DIMM DIMM
Interconnect
Low failure
rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
CPU CPU CPU CPU
31
Supportive Point 3: Unique Value-Added Components Eliminate I/O Bottlenecks and Significantly Increase I/O Performance
Data compression card iNIC
The data compression card applies to Hadoop and
WebServer services, reducing CPU usage and storage
space while improving service performance.
The CPU usage is lower than 20%.
The compression performance reaches 350 MB/s, and
decompression performance achieves 600 MB/s.
The iNIC is a high-performance network interface card. It
applies to firewall, network detection and analysis, and
virtualization applications.
The iNIC adopts load balancing among multiple queues
and zero copy to provide high performance and lower CPU
usage.
PCIe SSD card
Huawei is the first company which achieves
commercial use of PCIe SSD cards in the industry.
Dynamic wear leveling algorithm, bad block
management, and RAID 5 properties are embedded
to provide ultra high reliability.
The PCIe SSD card Improves SQL Server 2012
query performance by 1.5 to 4 times.
The PCIe SSD card reduces query time by 15 to
3000 times.
Low failure
rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
32
Supportive Point 4: Core Technologies of High-End Fault-Tolerant Servers
Four technological breakthroughs:
CPU interconnect chip
Fault-tolerant OS
Virtualization software
35 system reliability features Ceremony of online
application demonstration
2012: Huawei 32-socket, high-end, fault-tolerant servers
completed its online application demonstration on the PCRM
system of Agricultural Bank of China (ABC) and accounting
system of China Unicom.
2007: Huawei participated in the High-End Fault Tolerant
Computer Project of National High-tech R&D Program of
China (863 Program).
Low failure
rate
Outstanding energy
efficiency
Security and reliability
High availability
Continuous innovation
HUAWEI ENTERPRISE ICT SOLUTIONS A BETTER WAY
Copyright©2013 Huawei Technologies Co., Ltd. All Rights Reserved.
The information in this document may contain predictive statements including, without limitation, statements regarding the future financial and operating results, future product
portfolio, new technology, etc. There are a number of factors that could cause actual results and developments to differ materially from those expressed or implied in the predictive
statements. Therefore, such information is provided for reference purpose only and constitutes neither an offer nor an acceptance. Huawei may change the information at any time
without notice.
34
DEMT Introduction DEMT: Dynamic Energy Management Technology
The DEMT is a dynamic energy-saving technology developed by Huawei. The technology automatically monitors server
resource usage based on service loads, dynamically adjust the server running status based on resource utilization,
provides minimum power based on service requirements, and minimizes server power consumption. The DEMT reduces system power consumption for various load levels and does not affect service performance.
Energy saving is automatically triggered and monitored in closed-loop mode without manual intervention.
Server PSUs, CPUs, and fan modules can be adjusted to minimize power consumption, meeting different service requirements.
Energy-saving effect for DEMT in various service loads
Remarks: The data is obtained for the
RH2282 V2 server in the integer
computing benchmark test. The effect
varies for different server configurations.
35
Power Capping Technology Introduction Background The power supply and heat dissipation costs for data centers dramatically increase and exceed devices costs.
The power supply and heat dissipation capabilities of data centers do not meet requirements for service expansion.
The power supply and heat dissipation equipment of the IDC is underused.
Advantages Assigns power supply and heat dissipation resources based on actual device power, improving power efficiency.
Increases reliability without affecting service performance.
Eliminates the risks of insufficient heat dissipation and power supply if air conditioners are faulty or service loads
increase.
Supports more devices in a rack.
A rack can house more devices after the power
capping technology is implemented.
36
Energy Saving Design • Platinum PSUs are used, and the power conversion
efficiency reaches 94%
Dynamic load adjustment: evenly assigns PSU loads to ensure efficient
power conversion.
High-efficient 1+1 redundant PSUs, meeting system power supply
requirements
• Support for AC and 240 V high-voltage direct current
(HVDC) power supplies
Support for hot swap, meeting reliability requirements for 220 V AC and
240 V DC
Support for 240 V DC for 240 V HVDC equipment rooms, reducing energy
consumption by over 10%
PSU
240 V HVDC power supply
RH2285
Traditional UPS power supply and distribution
• Traditional UPS: Four times of DC/AC conversion. The power
supply efficiency for a single bus is about 64%. Two buses are
used in the equipment room, and the power supply efficiency is
less than 60%.
Support for AC
and 240 V HVDC
power supplies
• 240 V HVDC: One time of conversion reduced, power supply
efficiency up to 70%
• Efficiency is improved, and carriers' construction and operating
costs are greatly reduced.
240 V HVDC power supply and distribution
37
1. Failure Rate 15% Lower than the Average Industrial Level
Success Case: Tencent Data Center
The X6000 uses a passive backplane, no cables for internal components, and ventilation channels to ensure high
reliability.
The Dell C6220 has lower reliability than the X6000. The reasons as follows:
Uses an active backplane with low reliability
Uses a lot of cables to connect the backplane and server nodes, which leads to low cooling efficiency and a high failure rate. No
ventilation channel exists for the PSUs.
A large quantity of Huawei X series servers are used for Tencent's applications.
Huawei servers earn the reputation for the low failure rate and high reliability.
38
Competitiveness of High-Performance Servers
CPU/Memory TPS of a Physical Machine TPS of a VM Average
Performance Value
Performance Percentage
Lower than the Huawei
Server
Hardware Cost Percentage Lower
than Huawei Hardware (Different
Memory Configuration)
Huawei 4-socket
RH5885
4 x E7-4820/256 GB 2728 2952 2840
HP DL580 4 x E7-4820/128 GB About 2450 About 1800 2125 33.6% 10.90%
IBM X3850 4 x E7-4820/64 GB 2366 1149 1756 61.7% 17%
Inspur NF8560M2 4 x E7-4820/64 GB 1874 370 1122 153.1% 17%
Fujitsu Test failed (Only one group
of data can be tested.)
Test failed
4 sockets
CPU/Memory TPS of a Physical
Machine TPS of a VM
Average Performance
Value
Performance Percentage Lower than
the Huawei Server
Huawei 8-socket RH5885 8 x E7-8830/256 GB 4029 4250 4140
HP DL980 8 x E7-8830/256 GB
IBM X3950 8 x E7-8830/256 GB 3338 1903 2620 58.0%
Inspur TS850 621 466 543.5 661.7%
8 sockets
Remarks: high-end virtualization server comparison from China Mobile