# coding challenge profiler alltoallv 去看 mpi standard 我們要有一個 GIT HUB ### task 1 balance 的意思就是大家都傳一樣差不多的檔案或是所有rank都有用到 ### task 2 clone it to ur own github (own area) to improve WRF-already 3 domain GPAW- soon ### task 3 heavier - 呼叫的次數較多 他那一行的 CALL 應該是說 3次 比1 次多 color 那裏的單位是 BYTE to see which is heavier ### task 4 create the sum of patterns take each pattern * weight ### task 5 1- 0-4k is one colars 0-10 get one colar ### task 6 we found 2 node being use atav other not using 時間不一樣 WARF 沒用很多ATAV TIME不會差很多? ### task 7 ### task 8 ### .. output 下面那個 是RANK 的 EXAMPLE Hi all, we will start in few min Ophir Maor 對所有人說: 下午 11:00 We will wait another 2-3 min to start Ophir Maor 對所有人說: 下午 11:06 Feel free to post questions in the chat Alberto Garrote 對所有人說: 下午 11:33 what about the 4 to 4 text we saw in the pattern screen? Alberto Garrote 對所有人說: 下午 11:34 aah okay, thanks Susanne Malheiros 對所有人說: 下午 11:53 One question about the heatmap, what is the meaning of the diagnoal grid points? Does it mean that rank i is sending data to itself? Ophir Maor 對所有人說: 下午 11:55 yes, it is possible Susanne Malheiros 對所有人說: 下午 11:55 thank you! Luc-Henri Jolly 對所有人說: 下午 11:58 Does the profiler work when alltoallv is used with the MPI_IN_PLACE option ? Ophir Maor 對所有人說: 下午 11:59 We are not sure Ophir Maor 對所有人說: 下午 11:59 We expect it would work Luc-Henri Jolly 對所有人說: 上午 12:00 Ok, Thanks ! Luc-Henri Jolly 對所有人說: 上午 12:01 April 26Th Alexander Haller Heidelberg 對所有人說: 上午 12:01 Thank you Luc-Henri Jolly 對所有人說: 上午 12:01 Bye ! Alberto Garrote 對所有人說: 上午 12:01 bye Sami Haidar 對所有人說: 上午 12:01 Thanks! Ophir Maor 對所有人說: 上午 12:01 Thanks everyone for joining ## 重要的? main goal gather data to understand perfomance bottlencks to understand why in many case we use alltoallv end up with the perfomance we are expecting 就算我們不知道程式的source code也能追蹤程式 (就算是在不同平台) alltoallv其中一個重要的資訊 因為要一起溝通 所以會有先完成的核心會有等待的時間 因此呼叫會分成 late arrival 和 excution time pattern focus on 有多少ranks參與MPI alltoallv calls heat map focus on 多大的檔案在不同ranks傳輸 ###### tags: `coding challenge`