Á¤º¸°úÇÐȸ ³í¹®Áö A : ½Ã½ºÅÛ ¹× ÀÌ·Ð
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
Æù ³ëÀ̸¸ ±¸Á¶¸¦ äÅÃÇÑ µ¥ÀÌŸÇÃ·Î¿ì ´ÙÁß½º·¹µå ÄÄÇ»ÅÍ |
¿µ¹®Á¦¸ñ(English Title) |
Dataflow Multithreaded Computer Organization with von Neumann Architecture |
ÀúÀÚ(Author) |
±èÁ¤È¯
ÇÑ»ó¿µ
±è¼öÈ«
Junghwan Kim
Sangyong Han
Soohong Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 25 NO. 09 PP. 0917 ~ 0925 (1998. 09) |
Çѱ۳»¿ë (Korean Abstract) |
´ÙÁß½º·¹µå ¹æ½ÄÀº º´·Ä ó¸®ÀÇ ±âº» ¹®Á¦ÀÎ Áö¿¬ ½Ã°£°ú µ¿±âÈ ¹®Á¦ÀÇ ÇØ°á¿¡ È¿°úÀûÀÌ´Ù. ´ÙÁß½º·¹µå ÄÄÇ»ÅÍ DAVRID´Â µ¥ÀÌŸÇÃ·Î¿ì ¸ðµ¨¿¡ Æù ³ëÀ̸¸ ¸ðµ¨ÀÇ ÀåÁ¡À» Èí¼öÇÑ °è»ê¸ðµ¨À» ÃëÇÏ°í ÀÖ´Ù. º» ³í¹®¿¡¼´Â ¼º´É ºÐ¼®À» ÅëÇØ DAVRIDÀÇ ¹®Á¦Á¡µé Á¦½ÃÇÏ°í, »õ·Î¿î ÄÄÇ»ÅÍ ±¸Á¶ MVA(Dataflow Multithreaded Computer Organization with Von Neumann Architecture)¸¦ Á¦¾ÈÇÑ´Ù. DAVRID´Â °è»ê ¸ðµ¨ Ãø¸é¿¡¼ Áö¿¬ ½Ã°£°ú µ¿±âÈ ¹®Á¦ÀÇ ÇØ°á¿¡ ¸Å¿ì È¿°úÀûÀÌÁö¸¸, ±¸Á¶ÀûÀ¸·Î´Â Áö¿ª¼º È°¿ë¿¡ ¾î·Á¿òÀÌ ÀÖ´Ù. Á¦¾ÈµÈ MVA´Â DAVRIDÀÇ °è»ê ¸ðµ¨Àº ±×´ë·Î À¯ÁöÇÏ¸é¼ ±âÁ¸ÀÇ ÀϹÝÀûÀÎ Æù ³ëÀ̸¸ ±¸Á¶¸¦ äÅÃÇÔÀ¸·Î½á, È¿À²ÀûÀÎ º´·Ä 󸮸¦ ²ÒÇÏ°í ÀÖ´Ù. MVA¿¡ ´ëÇÑ ¼º´É ºÐ¼® °á°ú, DAVRID¿¡ ºñÇØ Ã³¸® ½Ã°£ÀÌ ´ëÆø °¨¼ÒÇÏ¿´À¸¸ç, °¢ ÀåÄ¡ÀÇ ÀÌ¿ë·üÀÌ Áõ°¡ÇÏ¸é¼ ÀÌ¿ë·ü ÆíÂ÷°¡ °¨¼ÒÇÏ¿´´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Multithreading is effective to solve the latency and the synchronization problem which are one of fundamental issues in parallel processing. One of multithreaded computers, DAVRID has the computational model which is based on the dataflow model together with the advantages of the von Neumann model. In this paper, we analyze some problems of the DAVRID through performance evaluation and propose a new architecture, called MVA(Dataflow Multithreaded Computer Organization with Von Neumann Architecture). The DAVRID effectively solves latency and synchronization problems from the view point of computational model, but has difficulty in exploiting locality from the view point of the architecture. The MVA pursuits efficient parallel processing in the way it inherits the computational model of the DAVRID and has conventional von Neumann architecture. Simulation results show that the execution time of the MVA is significantly reduced compared with that of the DAVRID, and the high variance of utilization among units is also reduced while overall utilization is increased.
|
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|