Ȩ > ÀڷḶ´ç > ÀÚ·á°Ë»ö > Ç¥ÁØ
ÀÚ·á °Ë»ö°á°ú
Ç¥ÁØÁ¾·ù | Á¤º¸Åë½Å´ÜüǥÁØ(TTAS) | ||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Ç¥ÁعøÈ£ | TTAR-10.0116 | ±¸ Ç¥ÁعøÈ£ | |||||||||||||||
Á¦°³Á¤ÀÏ | 2019-11-07 | ÃÑ ÆäÀÌÁö | 0 | ||||||||||||||
ÇÑ±Û Ç¥Áظí | x86±â¹Ý °í¼º´ÉÄÄÇ»Æà ¼¹ö ¼º´É ÃøÁ¤(±â¼úº¸°í¼) | ||||||||||||||||
¿µ¹® Ç¥Áظí | Measuring Performance of x86-based High-Performance Computing Servers (Technical Report) | ||||||||||||||||
ÇÑ±Û ³»¿ë¿ä¾à | 2017³â x86 ¼¹ö´Â Àü³â µ¿±â´ëºñ 46.8% ¼ºÀåÇÏ°í ÀÖÀ¸³ª, ±¹»ê °í¼º´ÉÄÄÇ»Æà ¼¹ö´Â ¿Ü»ê °í¼º´ÉÄÄÇ»Æà ¼¹ö ´ëºñ ÀúÆò°¡µÇ°í ÀÖ´Ù. À̸¦ °³¼±Çϱâ À§ÇÏ¿© ÀÌ ±â¼úº¸°í¼¿¡¼´Â °í¼º´ÉÄÄÇ»Æà ¼¹öÀÇ °´°üÀûÀÎ ¼º´ÉÀ» ÃøÁ¤ÇÒ ¼ö ÀÖ´Â ¹æ¹ýÀ» Á¦½ÃÇÏ°íÀÚ ÇÏ¸ç °´°üÀûÀÎ ¼º´ÉÀ» ÃøÁ¤Çϱâ À§ÇÏ¿© º¥Ä¡¸¶Å© µµ±¸ÀÎ HPL(High Performance Linpack)À» ÀÌ¿ëÇÑ´Ù. HPLÀº ´ÜÀÏ ¼¹öÀÇ ¼º´ÉÀ» ÃøÁ¤ÇÒ ¼ö ÀÖ´Â LINPACK º¥Ä¡¸¶Å©¸¦ º´·ÄÈÇÏ¿© ºÐ»ê½Ã½ºÅÛ¿¡¼ ¼öÇà °¡´ÉÇϵµ·Ï ±¸ÇöµÇ¾î ÀÖÀ¸¸ç, ÀÛ¾÷ÀÇ Å©±â, ÀÛ¾÷À» ¼öÇàÇÒ ³ëµå ¼ö µîÀ» ¼³Á¤ÇÏ¿© ºÐ»ê½Ã½ºÅÛÀÇ ¼º´ÉÀ» Flops(Floating point operation per second)¶ó´Â ¼º´É ÁöÇ¥¸¦ ÅëÇØ °´°üÀûÀÎ ¼öÄ¡·Î Ãâ·ÂÇÒ ¼ö ÀÖ´Ù. ¶ÇÇÑ Flops´Â ÃÊ´ç ºÎµ¿¼Ò¼öÁ¡ ¿¬»ê Ƚ¼ö¸¦ ÀǹÌÇϴµ¥, ºÎµ¿¼Ò¼öÁ¡ ¿¬»êÀº ÀϹÝÀûÀÎ Á¤¼ö ¿¬»ê¿¡ ºñÇØ º¹ÀâÇϱ⠶§¹®¿¡ °í¼º´É ÄÄÇ»ÅÍÀÇ CPU ¼º´É ÃøÁ¤ ÁöÇ¥·Î »ç¿ëµÈ´Ù. | ||||||||||||||||
¿µ¹® ³»¿ë¿ä¾à | In 2017 the high-performance computing(HPC) market mainly with x86 servers has grown by 46.8% more over the previous year, but the domestic HPC server industry has been inferior to the global ones. In the technical report to help the domestic industry at their imporving it. The technical report presents a practical information on benchmarkings that measures the objective performance of HPC servers in a cluster and HPL (High Performance Linpack) which is a benchmark tool for a clusterd HPC servers. HPL is implemented so that LINPACK benchmark that can measure the performance of a single server is parallelized and can be executed on a distributed system, setting the size of the work, the number of nodes executing the task, etc. It is possible to output the performance of the distributed system as an objective numerical value using the performance index, Flops. Flops means the number of floating point operations per second. Since floating point arithmetic is more complicated than general integer arithmetic, it is used as a CPU performance index of a HPC servers. | ||||||||||||||||
°ü·Ã IPR È®¾à¼ | Á¢¼öµÈ IPR È®¾à¼ ¾øÀ½ | ||||||||||||||||
°ü·ÃÆÄÀÏ | TTAR-10.0116.pdf | ||||||||||||||||
Ç¥ÁØÀÌ·Â |
|