Á¤º¸°úÇÐȸ ³í¹®Áö A : ½Ã½ºÅÛ ¹× ÀÌ·Ð
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
º´·Ä ½Ã½ºÅÛ È¯°æ ÇÏ¿¡¼ ºñÁ¤Çü ÀÀ¿ë ÇÁ·Î±×·¥À» À§ÇÑ ÀÔÃâ·Â ½Ã½ºÅÛÀÇ ¼³°è ¹× ±¸Çö |
¿µ¹®Á¦¸ñ(English Title) |
Design and Implementation of An I/O System for Irregular Applications under Parallel System Environments |
ÀúÀÚ(Author) |
³ëÀçÃá
¹Ú¼º¼ø
¾Ë·Ï »þ¿ìµå¸®
±Ç¿À¿µ
Jaechun No
Sung-Soon Park
Alok Choudhary
Oh-Young Kwon
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 26 NO. 11 PP. 1318 ~ 1332 (1999. 11) |
Çѱ۳»¿ë (Korean Abstract) |
º» ³í¹®¿¡¼´Â ÀÔÃâ·Â ÀÀ¿ëÀ» À§ÇØ collective I/O ±â¹ýÀ» ±â¹ÝÀ¸·Î ÇÑ ½ÇÇà½Ã°£ ½Ã½ºÅÛÀÇ ¼³°è, ±¸Çö ±×¸®°í ±× ¼º´ÉÆò°¡¸¦ ±â¼úÇÑ´Ù. ¿©±â¼´Â ¸ðµç ÇÁ·Î¼¼¼°¡ µ¿½Ã¿¡ I/O ¿ä±¸¿¡ µû¶ó ½ºÄÉÁ층Çϸç I/O¸¦ ¼öÇàÇÏ´Â collective I/O ¹æ¾È°ú ÇÁ·Î¼¼¼µéÀÌ ¿©·¯ ±×·ìÀ¸·Î ¹À̾î, ´ÙÀ½ ±×·ìÀÌ µ¥ÀÌÅ͸¦ Àç¹è¿ÇÏ´Â Åë½ÅÀ» ¼öÇàÇÏ´Â µ¿¾È ¿ÀÁ÷ ÇÑ ±×·ì¸¸ÀÌ µ¿½Ã¿¡ I/O¸¦ ¼öÇàÇÏ´Â pipelined collective I/O µîÀÇ µÎ °¡Áö ¼³°è¹æ¾ÈÀ» »ìÆ캻´Ù. Pipelined collective I/OÀÇ Àüü °úÁ¤Àº I/O ³ëµå Ãæµ¹À» µ¿ÀûÀ¸·Î ÁÙÀ̱â À§ÇØ ÆÄÀÌÇÁ¶óÀεȴÙ. ÀÌ»óÀÇ ¼³°è ºÎºÐ¿¡¼´Â µ¿ÀûÀ¸·Î Ãæµ¹ °ü¸®¸¦ À§ÇÑ Áö¿øÀ» Á¦°øÇÑ´Ù. º» ³í¹®¿¡¼´Â ´Ù¸¥ ³ëµåÀÇ ¸Þ¸ð¸® ¿µ¿ª¿¡ ÀÌ¹Ì Á¸ÀçÇÏ´Â µ¥ÀÌÅ͸¦ Àç »ç¿ëÇÏ¿© I/O ºñ¿ëÀ» ÁÙÀ̱â À§ÇØ collective I/O ¹æ¾È¿¡¼ÀÇ ¼ÒÇÁÆ®¿þ¾î ij½³ ¹æ¾È°ú µÎ °¡Áö ¸ðÇü¿¡¼ÀÇ chunking°ú ¿Â¶óÀÎ ¾ÐÃà¹æ¾ÈÀ» ±â¼úÇÑ´Ù. ±×¸®°í À̻󿡼 ±â¼úÇÑ ¹æ¾ÈµéÀÌ ÀÔÃâ·ÂÀ» À§ÇØ ³ôÀº ¼º´ÉÀ» º¸ÀÓÀ» ±â¼úÇϴµ¥, ÀÌ ¼º´É°á°ú´Â Intel Paragon°ú ASCI/Red teraflops ±â°è »ó¿¡¼ ½ÇÇèÇÑ °ÍÀÌ´Ù. ±× °á°ú ÀÀ¿ë ·¹º§¿¡¼ÀÇ bandwidth´Â peak point°¡ 55%±îÁö ÃøÁ¤µÇ¾ú´Ù. |
¿µ¹®³»¿ë (English Abstract) |
In this paper we present the design, implementation and evaluation of a runtime system based on collective I/O techniques for irregular applications. We present two design, namely, "Collective I/O" and "Pipelined Collective I/O". In the first scheme, all processors participate in the I/O simultaneously, making scheduling of I/O request simpler but creating a possibility of contention at the I/O nodes. In the second approach, processors are grouped into several groups, so that only one group performs I/O simultaneously, while the next group performs communication to rearrange data, and this entire process is pipelined to reduce I/O node contention dynamically. In other words, the design provides support for dynamic contention management. Then we present a software caching method using collective I/O to reduce I/O cost by reusing data already present in the memory of other nodes. Finally, chunking and on-line compression mechanisms are included in both models. We demonstrate that we can obtain significantly high-performance for I/O above what has been possible so far. The performance results are presented on an Intel Paragon and on the ASCI/Red teraflops machine. Application level I/O bandwidth up to 55% of the peak is observed. |
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|