• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ÇÐȸÁö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ÇÐȸÁö > µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

µ¥ÀÌÅͺ£À̽º ¿¬±¸È¸Áö(SIGDB)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) È¿À²ÀûÀÎ ºòµ¥ÀÌÅÍ Ã³¸®¸¦ À§ÇÑ °í¼Ó ºÐ»ê Ŭ·¯½ºÅÍ ±â¹Ý RDMA ¶óÀ̺귯¸®
¿µ¹®Á¦¸ñ(English Title) High-speed Distributed Cluster-based RDMA Library for Efficient Big Data Processing
ÀúÀÚ(Author) Á¤·¡¿ø   ±æ¸í¼±   ¹®¾ç¼¼   ÃÖÇüÁø   Laewon Jeong   Myeong-Seon Gil   Yang-Sae Moon   Hyung-Jin Choi  
¿ø¹®¼ö·Ïó(Citation) VOL 38 NO. 03 PP. 0035 ~ 0046 (2022. 12)
Çѱ۳»¿ë
(Korean Abstract)
º» ³í¹®¿¡¼­´Â ºÐ»ê Ŭ·¯½ºÅÍ¿¡ ÃÖÀûÈ­µÈ RDMA(Remote Direct Memory Access) ¶óÀ̺귯¸®¸¦ Á¦¾ÈÇÑ´Ù. RDMA´Â CPU¸¦ °ÅÄ¡Áö ¾Ê°í ³ëµå °£ ¸Þ¸ð¸®¿¡ Á÷Á¢ µ¥ÀÌÅ͸¦ ¼Û¼ö½ÅÇÏ´Â °í¼º´É ³×Æ®¿öÅ© ÇÁ·ÎÅäÄÝÀÌ´Ù. ±×·¯³ª, ±âÁ¸ RDMA °³¹ß ¹æ½ÄÀº º¹Àâµµ°¡ ¸Å¿ì ³ô°í, ÀÏ´ëÀÏ Åë½Å ±â¹Ý ¸ðµ¨·Î ÀÎÇØ ´Ù¼ö ³ëµå ȯ°æ¿¡ Àû¿ëÀÌ ¾î·Á¿î ¹®Á¦°¡ ÀÖ´Ù. À̸¦ ÇØ°áÇϱâ À§ÇØ, º» ³í¹®¿¡¼­´Â ±âÁ¸ API¸¦ ÀϹÝÈ­ÇÏ¿© ´Ù¼ö ³ëµå¿¡µµ ½±°Ô Àû¿ëÇÒ ¼ö ÀÖ´Â »õ·Î¿î RDMA ¶óÀ̺귯¸®¸¦ ¼³°èÇÏ°í ±¸ÇöÇÑ´Ù. Á¦¾È ¶óÀ̺귯¸®´Â 1) º¯¼ö ¼³Á¤, 2) Åë½Å Á¤º¸ »ý¼º, 3) RDMA ¼öÇà, 4) RDMA Á¾·áÀÇ ³× ´Ü°è·Î Åë½ÅÀ» ¼öÇàÇÑ´Ù. Á¦¾ÈÇÑ RDMA ¶óÀ̺귯¸®¸¦ »ç¿ëÇÑ Åë½Å ¸ðµâ ±¸Çö °á°ú, ´Ù¼ö ³ëµå °£ÀÇ µ¥ÀÌÅÍ ¼Û¼ö½ÅÀÌ Á¤»óÀûÀ¸·Î ¼öÇàµÊÀ» ½ÇÁ¦ ½ÇÇèÀ» ÅëÇØ È®ÀÎÇÏ¿´´Ù. °á°úÀûÀ¸·Î, Á¦¾È ¶óÀ̺귯¸®´Â ÁøÀÔ À庮ÀÌ ³ô¾Ò´ø RDMA ÀÀ¿ë °³¹ß ³­À̵µ¸¦ ³·Ãß°í, ±âÁ¸ ºÐ»ê ó¸® ȯ°æ¿¡µµ ½±°Ô È°¿ëÇÒ ¼ö ÀÖ´Â È¿À²ÀûÀÎ ¿¬±¸ °á°ú¶ó »ç·áµÈ´Ù.
¿µ¹®³»¿ë
(English Abstract)
In this paper, we deal with an RDMA (Remote Direct Memory Access) development technique optimized for distributed clusters. RDMA is a high-performance network protocol that directly transmits and receives data between nodes in memory bypassing a CPU. However, the existing RDMA development method has a very high complexity, and it is very difficult to apply to multi-node environments due to its one-to-one communication-based model. To solve this problem, in this paper we design and implement a new RDMA library that can be easily applied to multiple nodes by generalizing the existing API. The proposed library performs RDMA communications in four steps: 1) setting variables, 2) generating communication information, 3) executing RDMA, and 4) ending RDMA. Through the real communication module implementation using the proposed RDMA library, we confirm that data transmission and reception between multiple nodes operate correctly. We believe that our library can be an efficient way to reduce the difficulty of RDMA application development, which has high entry barriers, and be easily utilized in the existing distributed processing environments.
Å°¿öµå(Keyword) ÀÎÇǴϹêµå   Remote Direct Memory Access   °í¼Ó ºÐ»ê 󸮠  ºÐ»ê Ŭ·¯½ºÅÍ   ºòµ¥ÀÌÅÍ Ã³¸®   InfiniBand   High-speed distributed processing   Distributed cluster   Big data processing  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå