Á¤º¸°úÇÐȸ ³í¹®Áö C : ÄÄÇ»ÆÃÀÇ ½ÇÁ¦
Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
°¡»ó Ŭ·¯½ºÅÍ È¯°æ¿¡¼ ÇÏµÓ ¸Ê¸®µà½ºÀÇ ¼º´É Çâ»óÀ» À§ÇÑ ºÎÇÏºÐ»ê ±â¹ý |
¿µ¹®Á¦¸ñ(English Title) |
Load Balancing for Improving Hadoop MapReduce Performance in Virtual Cluster |
ÀúÀÚ(Author) |
Á¤´ë¿µ
³²À±¼º
À̱ǿë
¹Ú¼º¿ë
DaeYoung Jung
YoonSung Nam
KwonYong Lee
SungYong Park
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 19 NO. 12 PP. 0713 ~ 0717 (2013. 12) |
Çѱ۳»¿ë (Korean Abstract) |
°í¼º´É ÄÄÇ»Æà µîÀÇ ´Ù¾çÇÑ ºÐ¾ß¿¡¼ ³Î¸® ÀÌ¿ë µÇ´ø Ŭ·¯½ºÅÍ ÄÄÇ»ÆÃÀÌ Ãֱ٠Ŭ¶ó¿ìµå ¼ºñ½ºÀÇ µîÀåÀ¸·Î Ŭ¶ó¿ìµå »ó¿¡¼ °¡»ó Ŭ·¯½ºÅÍ·Î ±¸¼ºµÇ°í ÀÖ´Ù. ´ë¿ë·® µ¥ÀÌÅÍ Ã³¸®ÀÇ ´ëÇ¥Àû ºÐ»êó¸® Ç÷§ÆûÀÎ ÇÏµÓ Å¬·¯½ºÅÍÀÇ ±¸¼ºµµ Ŭ¶ó¿ìµå »óÀ¸·Î À̵¿ÇÏ´Â Ãß¼¼ÀÌ¸ç °¡»ó ¸Ê¸®µà½º Ŭ·¯½ºÅÍÀÇ ¼º´É Çâ»óÀ» À§ÇÑ ¿¬±¸°¡ È°¹ßÈ÷ ÀÌ·ç¾îÁö°í ÀÖ´Ù. ¸Ê¸®µà½º´Â ºÐ»ê 󸮵Ǵ ¸ðµç ŽºÅ©°¡ Á¾·áµÇ¾î¾ß ÃÖÁ¾ °á°ú¸¦ µµÃâÇÒ ¼ö Àִ Ư¼ºÀ» °®°í ÀÖÀ¸¹Ç·Î, ŽºÅ© µéÀÇ ¿Ï·á ½Ã°£ÀÌ ºÒ±ÕµîÇϸé Àüü ¸Ê¸®µà½ºÀÇ ¼º´ÉÀÌ Ç϶ôÇÑ´Ù´Â ´ÜÁ¡ÀÌ ÀÖ´Ù. ÇÏµÓ ¸Ê¸®µà½º¿¡¼´Â Ãß·ÐÀû ½ÇÇà ±â¹ýÀ» »ç¿ëÇÏ¿© ÀÌ ¹®Á¦¸¦ ÇØ°áÇÏ°íÀÚ ÇÏ¿´Áö¸¸ °¡»ó Ŭ·¯½ºÅÍ¿¡¼´Â Ŭ¶ó¿ìµå ÀÚ¿ø ³¶ºñ¿Í °°Àº ¹®Á¦¸¦ ¹ß»ý½ÃŲ´Ù. º» ³í¹®Àº Xen ±â¹ÝÀÇ Å¬¶ó¿ìµå »ó¿¡¼ ±¸¼ºµÈ °¡»ó ¸Ê¸®µà½º Ŭ·¯½ºÅÍ¿¡¼ Ãß·ÐÀû ½ÇÇà¿¡ ÀÇÇØ ¹ß»ýÇÏ´Â ¹®Á¦¸¦ ÇØ°áÇÏ´Â ºÎÇÏºÐ»ê ±â¹ýÀ» Á¦¾ÈÇÑ´Ù. Á¦¾ÈÇÏ´Â ºÎÇÏºÐ»ê ±â¹ýÀº Xen Å©·¹µ÷ ½ºÄÉÁÙ·¯¿Í ¸®´ª½º ½ºÄÉÁÙ·¯¸¦ ŽºÅ© ¼öÇà½Ã°£¿¡ µû¶ó µ¿ÀûÀ¸·Î Á¶ÀýÇÏ¿© ŽºÅ©ÀÇ ¼öÇà½Ã°£ ºÒ±ÕµîÀ» ÇؼÒÇÑ´Ù. ½ÇÇèÀ» ÅëÇØ Å½ºÅ©µéÀÇ ¼öÇà½Ã°£ÀÌ ±âÁ¸ÀÇ ÇÏµÓ ¸Ê¸®µà½º¿¡ ºñÇØ ±ÕµîÇÏ°Ô ÀÌ·ç¾îÁö°í ³«¿ÀÀÚ Å½ºÅ©ÀÇ ¹ß»ýÀ» ¹æÁöÇÏ¿© ¼º´ÉÀÌ Çâ»óµÊÀ» º¼ ¼ö ÀÖ¾ú´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Recently, as cloud computing service has been more popular, the clusters, widely used for high performance computing, are gradually configured as virtual clusters on the cloud environments. Hadoop MapReduce cluster, which is one of the representative distributed processing platforms, is also moved into the cloud, so that a lot of researches have been conducted to improve the performance of virtual MapReduce cluster. Since the MapReduce cannot complete a job until all the tasks are finished, unbalanced completion times of tasks result in performance degradation of the MapReduce. Even Hadoop MapReduce uses a speculative execution to solve this problem, it makes other problems including waste of cloud resources in the virtual cluster environments. In this paper, we propose a new load-balancing method to solve the problems occurred by the speculative execution of virtual MapReduce cluster running on the Xen-based clouds. The proposed method dynamically adjusts Xen credit scheduler and Linux scheduler based on the completion times of tasks, and thereby reduces the amount of unbalanced completion times of tasks. We evaluated the proposed method with the original Hadoop MapReduce, and concludes that our load balancing method improves the MapReduce performance by balancing the completion time of tasks and preventing the occurrence of straggler tasks.
|
Å°¿öµå(Keyword) |
Ŭ¶ó¿ìµå
°¡»ó Ŭ·¯½ºÅÍ
¸Ê¸®µà½º
Ãß·ÐÀû ½ÇÇà
ºÎÇϺлê
virtual cluster
MapReduce
speculative execution
load balancing
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|