omniture

화웨이의 AI Fabric 최고 수준의 성능으로 Tolly 테스트 통과

Tolly Group
2019-04-01 10:16 1,474

(푸저우 중국 2019년 4월 1일 PRNewswire=연합뉴스) Huawei China Eco-Partner Conference 2019에서 화웨이가 자사의 AI Fabric Intelligent and Lossless Data Center Network[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ] 솔루션이 국제적으로 권위 있는 테스트 기관 Tolly Group이 실시한 비교 테스트와 증명[https://e.huawei.com/en/material/onLineView?MaterialID=334f712b8f584d718d41e9c2ec172c67 ]을 통과했다고 발표했다. 테스트 결과, AI 시대를 지향하는 전형적인 데이터 센터 서비스 시나리오에서 화웨이의 AI Fabric[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ]은 업계의 다른 주류 공급업체가 출시한 네트워킹 솔루션보다 우수한 성능을 보였고, 지능적인 무손실 데이터 센터 네트워크의 새로운 시대를 이끌었다.

화웨이의 Global Industry Vision (GIV)에 따르면, 2025년에는 AI 보급률이 86%에 달할 전망이라고 한다. 결정을 내리고, 사업 모델과 에코시스템을 구축하며, 고객 경험을 재구성하기 위해 AI를 이용하는 것이 핵심적인 동력원으로 작용할 전망이라고 한다. 그뿐만 아니라 AI는 ICT 기반시설 변혁도 주도할 것으로 예상된다. 화웨이는 시대의 발전에 적응하고자 화웨이 커넥트(HUAWEI CONNECT) 2018에서 AI Fabric Intelligent and Lossless Data Center Network[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ] 솔루션의 출시를 주도했으며, 이를 통해 기업이 고성능 데이터 센터를 설립하도록 지원하고, AI 서비스의 상업적 이용을 가속화했다.

Tolly는 AI 시대에 데이터 센터가 당면하는 전형적인 세 가지 응용 시나리오(고성능 컴퓨팅(high-performance computing, HPC), 인공 지능(artificial intelligence, AI)과 머신 러닝(machine learning, ML), 분산 저장(distributed storage))에서 화웨이와 다른 주류 공급업체의 데이터 센터 전환망 솔루션을 테스트했다. 그리고 Intel MPI Benchmark (IMB) 모델과 TensorFlow 모델을 비롯해 100% 100GE와 100GE/25GE를 기반으로 하는 동일한 스파인-리프 네트워킹(spine-leaf networking) 환경에서 처리량, 패킷 손실률 및 지연시간 같은 성능 지표를 기준으로 이들 솔루션을 측정했다. 엄격한 테스트 끝에, 화웨이의 AI Fabric[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ]이 모든 성능 지표에서 가장 높은 점수를 획득하고, 평균보다 30% 높은 성적을 기록했다.

Comparison of image processing efficiency of computing nodes between Huawei AI Fabric and other vendors' networking solutions in the AI/ML scenario (Based on Tolly test report)
Comparison of image processing efficiency of computing nodes between Huawei AI Fabric and other vendors' networking solutions in the AI/ML scenario (Based on Tolly test report)

화웨이 AI Fabric Intelligent and Lossless Data Center Network[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ]는 업계 주도적인 CloudEngine 시리즈 데이터 센터 스위치들로 구성되며, 제로 패킷 손실, 짧은 지연시간 및 높은 처리량을 특징으로 하는 지능형 무손실 망을 지원하고자 독특한 iLossless(TM) 전환 알고리듬을 이용한다. 이것이 바로 화웨이 AI Fabric[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network]이 보여준 업계 최고 수준의 성능 뒤에 숨겨진 비밀이다.

Tolly Group 설립자 겸 CEO Kevin Tolly는 "철저한 테스트와 증명 작업 끝에 패킷 손실 제로, 짧은 지연시간 및 높은 처리량을 특징으로 하는 화웨이 AI Fabric[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ] 솔루션에서 깊은 인상을 받았다"라며 "고객은 테스트 보고서 전문을 읽고 이 솔루션의 가치를 확실하게 파악할 수 있다. 더욱 지능적인 스케줄링과 무손실 포워딩을 통해 이 솔루션이 AI 시대 데이터 센터 네트워크 건설을 주도할 것이라고 확신한다"라고 말했다.

화웨이 데이터 센터 네트워크 도메인 사장 Leon Wang은 "AI 시대에 데이터 센터 네트워크는 지능적인 무손실 망으로 변신해야 한다"라며 "자사의 솔루션은 주요 인터넷 기업에서 이미 상업적으로 사용되고 있다"고 밝혔다. 이어 그는 "Tolly가 자사의 AI Fabric[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ]에 대해 독자적이고 엄격한 테스트를 진행한 것은 기쁜 일"이라면서 "테스트 결과, 자사의 솔루션이 업계로부터 큰 인정을 받았다. 자사는 앞으로도 계속해서 R&D 투자와 기술 혁신을 늘리고, 데이터 통신 산업의 발전을 도모하며, 고객과 파트너들에게 주요 제품과 솔루션을 제공할 것"이라고 설명했다.

클라우드 데이터 센터 네트워크 건설 부문의 선두주자 중 하나인 화웨이의 데이터 센터 네트워크는 전 세계 7,800개 기업에서 이미 상업적으로 성공리에 사용 중이며, 금융, 인터넷 및 통신사 같은 산업에서 고객의 디지털 변혁을 촉진하고 있다. AI Fabric[https://e.huawei.com/en/solutions/business-needs/enterprise-network/data-center-network/ai-fabric-network ] 솔루션을 출시한 화웨이는 더욱 지능적인 무손실 데이터 센터 네트워크를 주도하고, AI 서비스의 만연한 이용을 촉진하며, 완전히 연결된 지능적인 세상을 위해 파트너들과 협력할 계획이다.

추가 정보는 화웨이 AI Fabric에 관한 Tolly 테스트 보고서를 다운로드 받아서 확인할 수 있다.

화웨이 vs 카시오 https://e.huawei.com/en/material/onLineView?MaterialID=334f712b8f584d718d41e9c2ec172c67

화웨이 vs Mellanox https://e.huawei.com/en/material/onLineView?MaterialID=642a12984b0d49548d0235dcbeda3a9d

사진: https://photos.prnasia.com/prnh/20190329/2418590-1

출처: 화웨이(Huawei)

출처: 화웨이
관련 링크: