Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)
ÇѱÛÁ¦¸ñ(Korean Title) |
µµÄ¿ ÄÁÅ×À̳ÊÀÇ ¸ÖƼÄÚ¾î CPU ´ë¿ªÆø ¿µÇ⠺м® |
¿µ¹®Á¦¸ñ(English Title) |
Analysis of Impact of Multi-core CPU Bandwidth in Docker Containers |
ÀúÀÚ(Author) |
ÀÓÀα¸
ÁøÇö¿í
Yin-Goo Yim
Hyun-Wook Jin
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 24 NO. 12 PP. 0675 ~ 0680 (2018. 12) |
Çѱ۳»¿ë (Korean Abstract) |
µµÄ¿´Â °æ·® °¡»óÈ ±â¼ú·Î¼ »ç¿ëÀÚ¿¡°Ô ÀÀ¿ëÀÇ »ý»ê¼º°ú °ü¸® ¹× ¹èÆ÷¿¡ ´ëÇÑ ºñ¿ëÀ» Å©°Ô °¨¼Ò½ÃŲ´Ù. ÀÌ·¯ÇÑ ÀÌÁ¡À» °¡Áø µµÄ¿´Â Ŭ¶ó¿ìµå, HPC ȯ°æ µî »ê¾÷ ¹× ¿¬±¸ÀÇ ¿©·¯ ºÐ¾ß¿¡¼ È°¿ëµÇ¸ç, µ¿½Ã¿¡ °Ý¸®µÈ ÀÚ¿øÀ» È¿À²ÀûÀ¸·Î È°¿ëÇϱâ À§ÇÑ Áö¼ÓÀûÀÎ ¿¬±¸°¡ ÁøÇàµÇ°í ÀÖ´Ù. µµÄ¿´Â cgroupÀÇ ÀÚ¿ø°ü¸® ±â¹ýÀ» È°¿ëÇÑ´Ù. cgroupÀÇ CFS ½ºÄÉÁÙ·¯´Â °øÀ¯ºñÀ² ¹æ½Ä°ú ´ë¿ªÆø ¹æ½ÄÀÇ ÀÚ¿ø ¿¹¾à ¹æ½ÄÀ» Á¦°øÇÑ´Ù. ¶ÇÇÑ, ÄÚ¾î ģȵµ ±¸¼ºÀ» ÅëÇØ ÄÁÅ×À̳ʰ¡ µ¿ÀÛÇÏ´Â CPU ÁýÇÕÀ» ÁöÁ¤ÇÒ ¼ö ÀÖ´Ù. ÇÏÁö¸¸ »ç¿ëÀÚ ÀÔÀå¿¡¼ °øÀ¯ºñÀ² ¹æ½Ä¿¡ ºñÇØ ´ë¿ªÆøÀ» ±¸¼ºÇÏ´Â ÁÖ±â¿Í Çã¿ë·® ¹× ÄÚ¾î ģȵµ°¡ ÀÀ¿ëÀÇ ¼º´É¿¡ ¹ÌÄ¡´Â ¿µÇâÀº Å©°Ô °í·ÁµÇÁö ¾Ê´Â´Ù. º» ³í¹®¿¡¼´Â ÁÖ±â, Çã¿ë·®, ÄÚ¾î ģȵµ ±¸¼ºÀÌ ÄÁÅ×ÀÌ³Ê ÀÀ¿ëÀÇ ¼º´É¿¡ ¹ÌÄ¡´Â ¿µÇâÀ» ºÐ¼®ÇÑ´Ù. ½ÇÇèÀº ¹èÄ¡ ¿öÅ©·Îµå¿Í, ´ëÈÇü ¿öÅ©·Îµå¿¡¼ °¢°¢ ¼öÇàµÇ¾úÀ¸¸ç, ºÐ¼® °á°ú¸¦ ÅëÇØ ÄÁÅ×À̳ÊÀÇ ½ºÄÉÁÙ¸µ ÁÖ±â¿Í Çã¿ë·®, ÄÚ¾î ģȵµ ±¸¼ºÀ» ÅëÇÑ ÀÀ¿ë ¼º´É ÃÖÀûÈ °¡´É¼ºÀ» Á¦¾ÈÇÑ´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Docker is a lightweight virtualization technology that significantly reduces the cost of productivity and administration and deployment of applications. These benefits are exploited in many areas of industry and research, including cloud and HPC environments, and ongoing research is underway to efficiently use isolated resources. The docker uses cgroup resource management techniques. The cgroup¡¯s CFS scheduler provides a resource-reservation scheme, with a shares scheme and a bandwidth scheme. In addition, a cgroup can specify a set of CPUs, wherein the container operates through core affinity configuration. However, influence of the period and quota configuring bandwidth, core affinity on the performance of the application, is not considered as much as the shares scheme. In this paper, we analyze effects of period, quota, and core affinity on performance of container applications. Experiments were conducted on batch workloads and interactive workloads, respectively, and suggest the possibility of optimizing application performance of containers through analysis results.
|
Å°¿öµå(Keyword) |
µµÄ¿
ÄÁÅ×À̳Ê
cgroup
¼º´É ºÐ¼®
º¥Ä¡¸¶Å©
docker
container
cgroup
performance evaluation
benchmark testing
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|