High Throughput Computing Data Center Architecture
Zhulin (Zane) Wei, Shujie Zhang, Yuangang (Eric) Wang, Qinfen (Jeff) Hao, Guanyu Zhu, Junfeng Zhao, Haibin (Benjamin) Wang, Xi Tan (Jake Tam), Jian Li*, etc
Shannon Lab
Details see white paper in ISCA package; Contacts for further brainstorming: Jake Tam (tanxi@huawei.com) or Eric Wang ([email protected]) HUAWEI TECHNOLOGIES CO., LTD. www.huawei.com Outline
Big Data Challenges to Data Center Architecture
Huawei’s Vision on Future Data Center
DC 3.0: HTC-DC
Page 2 HUAWEI TECHNOLOGIES CO., LTD. 2012~present Evolution of Human Society 1971~2011
1764~1970 1971~2011 Post-Info Society： 4000 BC~1763 Avg. consumption of Info-Data per Capita Information Society： 2012 (Person Per Year) ** Avg. Internet access/per Capita Industrial Society： US：1960GB Avg. consumption of Electricity EU：1930GB per Capita 2011 (Every 100 Person) * US：78 CN：186GB Agricultural Society： 1970（kWh） EU：72 Avg. consumption of Protein per ： Capita US 723 CN：38 EU：2888 Protein consumption CN：151(1971)* as a identity for the development of civilization * Source: http://data.worldbank.org ** Source: IDC, 2012
Page 3 HUAWEI TECHNOLOGIES CO., LTD. Big Data Era: Enjoy Life via Data Consumption
PB level → Enjoy Intelligent Life
TB level → Enter the Age of Data
All Related Data
Intensity-related Data GB level → Obtain Real-time Info Cleaned Data
Page 4 HUAWEI TECHNOLOGIES CO., LTD. Big Data Challenges to Data Centers
Limitations of Current DC
• Data processing • Typically • Limited flexibility • High speed copper • Lower power capability Utilization<30% for deployment and interconnect efficiency • I/O bottleneck • Virtualization configuration • DC-level large- with high overhead • Complex operations scaled interconnect
Throughput Resource Utilization Management Scalability Energy Efficiency
• New medium • Resource • Intelligent • Optics based • New architecture • New architecture disaggregation Management interconnect for energy • New access • On-demand and • Self-healing efficient Mechanism flexible resource • Self-configuration computing allocation • Software-defined
Strategies
Page 5 HUAWEI TECHNOLOGIES CO., LTD. Evolution of Data Center Architecture
Full-stack elasticity
Disaggregation
Page 6 HUAWEI TECHNOLOGIES CO., LTD. Outline
Big Data Challenges to Data Center Architecture
Huawei’s Vision on Future Data Center
DC 3.0: HTC-DC
Page 7 HUAWEI TECHNOLOGIES CO., LTD. Huawei’s Vision on Future Data Centers
Resource Disaggregated
Page 8 HUAWEI TECHNOLOGIES CO., LTD. Big Data Oriented Architecture
Generator Revolution， IT System Revolution， ending the bottleneck of power generating ending the bottleneck of data processing Magnetic lines of force Coil CPU CPU CPU
Petabyte CPU CPU Level Data
New Coil Data Excitation Architecture CPU CPU power supply CPU Fixed coil generator: MWH PB Data-centric system:
break through the scale limitation of coil break through the bottleneck of data throughput
Rotating coil PB TB GB
Old CPU Magnetic field Computing Carbon Becket Architecture MEM brush DISK TAPE Conventional generator: KWH TB Computing-centric system: scale limitation of coil limitation of data throughput
Page 9 HUAWEI TECHNOLOGIES CO., LTD. Intelligent Management
Data Infrastructure Analysis
Application Analytics & Optimization
Management Infrastructure Analysis
Page 10 HUAWEI TECHNOLOGIES CO., LTD. Open, Standard based and Flexible Service Layer
So much more Generates Service
More Confusion content Frustration Feeling of being More More User overwhelmed devices apps Experience
More people Esp. young people
Page 11 HUAWEI TECHNOLOGIES CO., LTD. Adaptation for Task Variation
Data Center Adapts to different application and workload like a chameleon
Page 12 HUAWEI TECHNOLOGIES CO., LTD. High Scalability
Scale up easily Have ability can run between peak pressure and usual time
Page 13 HUAWEI TECHNOLOGIES CO., LTD. Green
Traditional data center consume power Build Green data center much
Page 14 HUAWEI TECHNOLOGIES CO., LTD. Outline
Big Data Challenges to Data Center Architecture
Huawei’s Vision on Future Data Center
DC 3.0: HTC-DC
Page 15 HUAWEI TECHNOLOGIES CO., LTD. HTC-DC Architecture Data Plane Management Plane DC OS App App App App DC Management Interface
Hadoop / JVM Resource Management Center
Dedicated OS Management SLA Self-Learning DC Group Job Scheduling Priority Load Balancing Partitioning …
Dedicated OS Partition / Light-Hypervisor Hardware HTC Programming Framework Information Interconnect CPU Memory I/O Abstract Resource Pool Pool Pool
Hardware
I/O Extensions
GPU Network Storage
Interconnects Computing Nodes Memory Nodes I/O Nodes
Page 16 HUAWEI TECHNOLOGIES CO., LTD. Hardware Architecture of HTC-DC
Page 17 HUAWEI TECHNOLOGIES CO., LTD. HTC-DC Key Feature
Resource Disaggregated Pooled DC -Level Resource Efficient Access Programming Protocol Framework (PRAP) HTC- DC Key Feature
Many-Core Optical Processor Interconnect
NVM Based Storage
Page 18 HUAWEI TECHNOLOGIES CO., LTD. Resource Disaggregated
On-demand Resource Allocation Based on Disaggregation
Page 19 HUAWEI TECHNOLOGIES CO., LTD. Pooled Resource Access Protocol (PRAP)
Lossless Transmission
MEM
CCPPU
PRAP
I/O
Simplified Data Packet Format Address Based Routing
Page 20 HUAWEI TECHNOLOGIES CO., LTD. Many-Core Processor
Page 21 HUAWEI TECHNOLOGIES CO., LTD. Summary
HTC-DC ：A new, green and intelligent DC 3.0 for Big Data era
HTC-DC is a new generation of DC architecture: Resource disaggregated Pooled Resource Access Protocol (PRAP) Many-core processors NVM as Storage Optical interconnects DC-level efficient programming framework
PB-level data processing, intelligent management, high scalability, and green
We welcome brainstorming, critics, suggestions and comments
Page 22 HUAWEI TECHNOLOGIES CO., LTD. High Throughput Computing Data Center Architecture
Zhulin (Zane) Wei, Shujie Zhang, Yuangang (Eric) Wang, Qinfen (Jeff) Hao, Guanyu Zhu, Junfeng Zhao, Haibin (Benjamin) Wang, Xi Tan (Jake Tam), Jian Li*, etc
Shannon Lab
Details see white paper in ISCA package; Contacts for further brainstorming: Jake Tam ([email protected]) or Eric Wang ([email protected]) HUAWEI TECHNOLOGIES CO., LTD. www.huawei.com Thank you
www.huawei.com
Copyright©2014 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.
Page 24 HUAWEI TECHNOLOGIES CO., LTD.