Welcome to NCLab

NETWORK and COMPUTING LABORATORY

1443188170_network-server

Super Computing

1443188152_car

Mobile Edge Computing

1443188157_Database-Cloud

eXplainable AI platform

1443188157_Database-Cloud

Crowdsourcing based Smart City

1443188157_Database-Cloud

Integrated Deep Learning Engine

1443188157_Database-Cloud

FPGA Virtualization

Network and Computing Laboratory(NCL)

1443188170_network-server

Super Computing

Supercomputing technology is efficiently operated to give high-tech research environments to local scientists, consequently enhancing science and technology competitiveness.

1443188157_Database-Cloud

Mobile Edge Computing

Mobile edge computing is a new computing paradigm that enables cloud computing capabilities at the edge of the cellular network, so as to reduce overhead of remote cloud access.

1443188152_car

XAI based Energy Platform

XAI, eXplainable AI is the most challenging area in the field of artificial intelligence which understands the process and explains the consequences. We are researching the energy platform for XAI in cooperation with KEPCO.

1443188152_car

Crowdsourcing based Smart City

With the development of IoT, a large amount of data collectd from citizens, devices and buildings should be analyzed to monitor and manage traffic, power plants and waste management etc.

1443188170_network-server

Integrated Deep Learning Engine

Deep learning is one of popular issues in the machine learning field. We are researching accelerated deep learning and developing a framework for distributed training.

1443188170_network-server

FPGA Virtualization

FPGA is getting more important on numerous research fields due to its design flexibility and power efficiency. Yet, FPGA is costly when it comes to frequent reprogramming. We introduce FPGA virtualization technique to overcome FPGA’s lengthy reprogramming.

Welcome to NCLab.

  • heterogeneous many-core hardware system for UHPC

Welcome to NCLab.

Distributed computing systems have become pervasive. From clusters to internet-worked computers, to mobile machines, distributed systems are being used to support a wide variety of applications. Our research works are focusing on key technologies underlying the design and engineering of distributed
computing systems and computing middleware.

Network and Computing Lab (NCL), a founder of Grid Middleware Research Center, has been focusing on advanced computing middleware and development of service management system in advanced networks, e.g. next generation network and future Internet. Especially, Grid PQRM (Policy Quorum-based Resource Management) developed through ITRC project for the past 4 years was evaluated as one of the best research projects by the government, Ministry of Education, Science and Technology in 2008. And, we have been developed Nano-Sensor Integrated Micro-Computing (NSIMC) applicable for health care system, which was a sort of biocomputing system for the identification of metabolic mechanism using the biochemistry index of human cells. The NSIMC provides the clinical estimation of the patient’s metabolic syndrome using energy-circulation model in the cell mitochondria and it also provides the collaborative environment for medical doctors, scientists and specialists. As for biocomputing, we developed an e-Organ simulation system that was utilizing Cyber Computing for in silico drug discovery to identify new drugs effectively from metadata of chemical compounds. This system provided high-performance computing for experiments of drug discovery with coordination and efficient execution management of geographically distributed complex applications. Furthermore, this system helped researchers share multidisciplinary computing, collaboration and simulation results.

NCL’s main research areas are cooperative high-performance computing platform (Cloud, IoT, TaaS, Connected Car), Deep Learning Platform, HPC interconnection and Integrated Service Broker Middleware managing computing resources.

Currently, we are developing 1) Deep Learning HPC Platform for high-speed processing, 2) V2C Cloud Computing system environment for smart car, 3) Giga Media based platform for Tele-Experience service software and 4) foundation techniques for developing national supercomputer

 

Supercomputing

 

 

Supercomputing Technology provides efficient high-tech research environments to local scientists. Consequently super computing technology has been an important area for enhancing science and technology competitiveness. As growing attention to some state of the art technologies requiring huge computing resources such as big data, machine learning, supercomputing technology is becoming more important as a foundation technology to accelerate them. We are researching not only on some techniques related to implement distributed systems for deep learning using Tensorflow but also on some foundation techniques. [more info]

●Interconnection Network in HPC

Network bottlenecks in HPC frequently dominate the performance degradations caused by CPU, cache and memory. As a scale of HPC system become larger, interconnect systems require efficient design of its composition. In addition, a cost charges for interconnect system approaches 25% of HPC construction cost. Some conditions for optimization are required in efficient interconnect system. 1) Minimize the degree of a regular network in order to get a cheap hardware cost, 2) Minimize the network diameter in order to get short paths used by the communication network (low latency) and 3) Maximize the network dimension to increase the number of nodes/processors for handling larger volumes of data (scalability). There are three design elements (topology, flow control and router) to decide the interconnect characteristics. Our approaches include Silicon photonics based high speed on-board router, pseudo-global adaptive routing method and also new emerging topology management. We expect better throughput, scalability, latency and energy consumption characteristics in new HPC interconnect system.

●Developing HPC (High Performance Computing) system for large scale deep learning acceleration

we are researching on developing distributed/parallel processing platform for large-scale deep learning. Through research on large-scale distributed processing platform such as caffe and tensorflow, we are developing DL-MDL which is multi layer interface and standard language.

keyword: supercomputing, Interconnection Network

Integrated Deep Learning Engine

 

 

Deep learning has recently received worldwide attention as a way to solve problems using machine learning. The size of the model is growing as it is found that larger models can provide better performance. Training large models, however, not only requires a large amount of memory, but also requires a lot of training time. To deal with this problem, training large-scale model in distributed environment has emerged as one of the most widely used solutions. We are researching on distributed computing platforms for efficient large-scale deep learning model training.

● As the size of the deep learning model grows, it is necessary to study accelerated learning of the deep learning model
● Development of HPC system for high-speed deep processing
● Define Deep Learning Model Description Language (DL-MDL) as an interface to integrate existing deep learning frameworks (eg, TensorFlow, Caffe, CNTK)
● Create a code that enables deep learning model written in DL-MDL to be trained in various deep learning frameworks
● Allocates resources suitable for the training among all the available cluster resources
● Automatically performs optimal model parallelism when using allocated resources
[more info]

keyword: deep learning, distributed system, Artificial Intelligence

Mobile Edge Computing

 

 

Along with 5G technology, Mobile Edge Computing (MEC) is attracting attention as one of the promising technologies to realize ultra reliable low latency communication (URLLC) services. The MEC places cloud-computing capabilities on the radio access network (RAN), which gets closer to end users. In this project, we focus on implementing the MEC prototype with hardware accelerators, especially FPGA to reduce the response time of an application located at the MEC. To do that, we are focusing on (1) Framework enabling FPGA-MEC support, (2) FPGA virtualization, and (3) Showing the PoC for MEC with FPGA acceleration
[more info]

keyword: Mobile Edge Comuting, URLLC

FPGA Virtualization

 

 

Field Programmable Gate Array(FPGA) is a hardware which allows users to build custom hardware structures of their own. Due to this Aspect, FPGA can provide almost infinite kinds of services and the usage on FPGA is growing on various research areas. In exchange for this flexibility however, one must suffer lengthy reprogramming to upload a custom design. To overcome this problem and maximize the efficiency of FPGA, we introduce FPGA virtualization via partial reconfiguration technique.
[more info]

keyword: Virtualization, Field Programmable Gate Array, FPGA

XAI based Energy Platform

 

 

Real-time analysis of power monitoring data streams is required to detect electrical equipment failure and response to that failure; failure to respond immediately could lead to major accidents such as blackout and fire. Real-time detection of power patterns that have high probability of failure from data stream continuously collected by the real-time monitoring instrument & distributed task execution technology for deep-learning-based Complex Event of time series data steam. Deep learning processing acceleration technology using real-time dynamic task allocation in a heterogeneous acceleration environment
[more info]

keyword: Explainable AI, Big Data Analysis, Energy Platform, KEPCO, Deep Learning Streaming

Crowdsourcing based Smart City

 

 

As the various and many mobile devices come into wide use, it becomes possible to improve the quality of life for citizens by providing the more smart service through being reported the local information of each mobile and using them. In the crowdsourcing based road environment report & local event map service framework, the illegal parking re-identification to solve the traffic congestion smartly and the local event map service that can provide local information to citizens in real time is provided.
[more info]

keyword: Smart city, Crowdsourcing, Semantic Aware, Local Event Map

Supercomputing

 

 

Supercomputing Technology provides efficient high-tech research environments to local scientists. Consequently suepr computing technology has been an important area for enhancing science and technology competitiveness. As growing attention to some state of the art technologies requiring huge computing resources such as big data, machine learning, supercomputing technology is becoming more important as a foundation technology to accelerate them. We are researching not only on some techniques related to implement distributed systems for deep learning using Tensorflow but also on some foundation techniques. [more info]

– Interconnection Network in HPC

Network bottlenecks in HPC frequently dominate the performance degradations caused by CPU, cache and memory. As a scale of HPC system become larger, interconnect systems require efficient design of its composition. In addition, a cost charges for interconnect system approaches 25% of HPC construction cost. Some conditions for optimization are required in efficient interconnect system. 1) Minimize the degree of a regular network in order to get a cheap hardware cost, 2) Minimize the network diameter in order to get short paths used by the communication network (low latency) and 3) Maximize the network dimension to increase the number of nodes/processors for handling larger volumes of data (scalability). There are three design elements (topology, flow control and router) to decide the interconnect characteristics. Our approaches include Silicon photonics based high speed on-board router, pseudo-global adaptive routing method and also new emerging topology management. We expect better throughput, scalability, latency and energy consumption characteristics in new HPC interconnect system.

– Developing HPC (High Performance Computing) system for large scale deep learning acceleration

we are researching on developing distributed/parallel processing platform for large-scale deep learning. Through research on large-scale distributed processing platform such as caffe and tensorflow, we are developing DL-MDL which is multi layer interface and standard language.

keyword: supercomputing, deep learning, distributed system, HPC, Artificial Intelligence

Mobile Edge Computing

 

 

Along with 5G technology, Mobile Edge Computing (MEC) is attracting attention as one of the promising technologies to realize ultra reliable low latency communication (URLLC) services. The MEC places cloud-computing capabilities on the radio access network (RAN), which gets closer to end users. In this project, we focus on implementing the MEC prototype with hardware accelerators, especially FPGA to reduce the response time of an application located at the MEC. To do that, we are focusing on (1) Framework enabling FPGA-MEC support, (2) FPGA virtualization, and (3) Showing the PoC for MEC with FPGA acceleration
[more info]

keyword: Mobile Edge Comuting, URLLC

Integrated Deep Learning Engine

 

 

Deep learning has recently received worldwide attention as a way to solve problems using machine learning. The size of the model is growing as it is found that larger models can provide better performance. Training large models, however, not only requires a large amount of memory, but also requires a lot of training time. To deal with this problem, training large-scale model in distributed environment has emerged as one of the most widely used solutions. We are researching on distributed computing platforms for efficient large-scale deep learning model training.
[more info]

keyword: deep learning, distributed system, Artificial Intelligence

FPGA Virtualization

 

 

Field Programmable Gate Array(FPGA) is a hardware which allows users to build custom hardware structures of their own. Due to this Aspect, FPGA can provide almost infinite kinds of services and the usage on FPGA is growing on various research areas. In exchange for this flexibility however, one must suffer lengthy reprogramming to upload a custom design. To overcome this problem and maximize the efficiency of FPGA, we introduce FPGA virtualization via partial reconfiguration technique.
[more info]

keyword: Virtualization, Field Programmable Gate Array, FPGA

Crowdsourcing based Smart City

As the various and many mobile devices come into wide use, it becomes possible to improve the quality of life for citizens by providing the more smart service through being reported the local information of each mobile and using them. In the crowdsourcing based road environment report & local event map service framework, the illegal parking re-identification to solve the traffic congestion smartly and the local event map service that can provide local information to citizens in real time is provided.
[more info]

keyword: Smart city, Crowdsourcing, Semantic Aware, Local Event Map

XAI based Energy Platform

 

Real-time analysis of power monitoring data streams is required to detect electrical equipment failure and response to that failure; failure to respond immediately could lead to major accidents such as blackout and fire. Real-time detection of power patterns that have high probability of failure from data stream continuously collected by the real-time monitoring instrument & distributed task execution technology for deep-learning-based Complex Event of time series data steam. Deep learning processing acceleration technology using real-time dynamic task allocation in a heterogeneous acceleration environment
[more info]

keyword: eXplainable AI, Big Data Analysis, Energy Platform, Deep Learning Streaming, KEPCO

MOBILE VERSION


373-1, Guseong-dong, Yuseong-gu Daejeon, 305-701, Korea
Phone: +82-42-350-3495 FAX: +82-42-350-7260
Network and Computing Lab http://ncl.kaist.ac.kr
( Webmaster: iop851@kaist.ac.kr ─ Last update: 2018.09.27 )


Last update: 2017.09.11
Webmaster: iop851@kaist.ac.kr