¿ËÈÕ£¬£¬£¬£¬£¬£¬£¬£¬ÓÉÖйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿Ôº×÷ΪµÚÒ»µ¥Î»Íê³ÉµÄ×ÔÁ¦×ÔÖ÷Ñо¿Ð§¹û¡¶LEVELLER:
Fair Communication Scheduling via Progress-Rate Awareness in Multi-Tenant
Training Clusters¡·±»¹ú¼ÊÅÌËã»úÍøÂçÁìÓò¶¥¼¶¾Û»áACM SIGCOMM£¨ACM Special Interest Group on Data Communication Conference£©2026ÕýʽÈÎÃü¡£¡£¡£¡£¡£¡£¡£¡£¸ÃЧ¹ûʵÏÖÁËÑëÆóÒÔµÚÒ»µ¥Î»Éí·ÝÔÚACM SIGCOMM½ÒÏþ×ÔÁ¦×ÔÑÐÂÛÎĵÄÀúÊ·ÐÔÍ»ÆÆ£¬£¬£¬£¬£¬£¬£¬£¬±ê¼Ç×ÅÖйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÚÔÆÅÌËãÍøÂç»ù´¡Ñо¿ÓëÏÏû³Á¢Òì·½ÃæÈ¡µÃÖ÷ҪϣÍû¡£¡£¡£¡£¡£¡£¡£¡£
ACM SIGCOMMÊÇÅÌËã»úÍøÂçÁìÓò×î¾ßÓ°ÏìÁ¦µÄ¹ú¼Ê¶¥¼¶Ñ§Êõ¾Û»áÖ®Ò»£¬£¬£¬£¬£¬£¬£¬£¬±»ÖйúÅÌËã»úѧ»áÍÆ¼öĿ¼ÁÐΪCCF AÀà¾Û»á¡£¡£¡£¡£¡£¡£¡£¡£ÎåÊ®¶àÄêÀ´£¬£¬£¬£¬£¬£¬£¬£¬SIGCOMM½ÒÏþµÄÖî¶à¾µäÑо¿Ò»Á¬Íƶ¯Êý¾ÝͨѶϵͳ½á¹¹¡¢ÍøÂçÐÒé¡¢Êý¾ÝÖÐÐÄÍøÂçºÍ»¥ÁªÍø»ù´¡ÉèÊ©µÄÑݽø£¬£¬£¬£¬£¬£¬£¬£¬Éî¿ÌÓ°ÏìÁËÏȽøÍøÂçÊÖÒÕµÄÉú³¤Æ«Ïò¡£¡£¡£¡£¡£¡£¡£¡£SIGCOMM¶ÔÂÛÎÄÖÊÁ¿ÒªÇ󼫸ߣ¬£¬£¬£¬£¬£¬£¬£¬Ç¿µ÷»ù´¡ÐÔТ˳¡¢Ç°Õ°ÐÔÓ°ÏìÓë¼áʵµÄϵͳʵÏÖ£¬£¬£¬£¬£¬£¬£¬£¬Â¼È¡Âʺã¾Ã´¦ÓڽϵÍˮƽ£¬£¬£¬£¬£¬£¬£¬£¬½üÄêÀ´Í¨³£Ô¼Îª16%¡£¡£¡£¡£¡£¡£¡£¡£ÆäÈÎÃüÂÛÎÄÍùÍùÊܵ½Ñ§Êõ½çÓ빤ҵ½çµÄÆÕ±é¹Ø×¢£¬£¬£¬£¬£¬£¬£¬£¬¶ÔÁ¢ÒìÊÖÒÕÂ䵨ºÍ¹¤ÒµÉú³¤¾ßÓÐÖ÷ÒªÍÆÐж¯Óᣡ£¡£¡£¡£¡£¡£¡£¾Ýͳ¼Æ£¬£¬£¬£¬£¬£¬£¬£¬×èÖ¹2025Ä꣬£¬£¬£¬£¬£¬£¬£¬ÉÐδÓÐÑëÆóÒÔµÚÒ»µ¥Î»Éí·ÝÔÚACM SIGCOMMÉϽÒÏþ×ÔÁ¦×ÔÖ÷Ñз¢Ð§¹û¡£¡£¡£¡£¡£¡£¡£¡£´Ë´ÎÖйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿ÔºÒÔµÚÒ»µ¥Î»Éí·Ý£¬£¬£¬£¬£¬£¬£¬£¬½«ÍêÈ«×ÔÁ¦×ÔÖ÷Ñз¢µÄЧ¹ûдÈëÕâÒ»¹ú¼Ê¶¥¼¶¾Û»á£¬£¬£¬£¬£¬£¬£¬£¬ÊµÏÖÁËÑëÆóÔÚÅÌËã»úÍøÂç¶¥¼¶Ñ§ÊõÎę̀ÉϵÄÀúÊ·ÐÔÍ»ÆÆ¡£¡£¡£¡£¡£¡£¡£¡£

ͼ£ºLEVELLERÉè¼Æ¼Ü¹¹£¨ÉÏ£©ÓëЧ¹û£¨Ï£©
±¾´Î±»ÊÕ¼µÄÂÛÎÄ¡¶LEVELLER: Fair Communication
Scheduling via Progress-Rate Awareness in Multi-Tenant Training Clusters¡·£¬£¬£¬£¬£¬£¬£¬£¬½â¾öGPU¼¯ÈºÖжà×⻧ͨѶ¹«ÕýÐÔÄÑÌâ¡£¡£¡£¡£¡£¡£¡£¡£¸ÃÊÂÇéÓÉÖйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿Ôº×ÊÉîÖ÷ÈÎÑо¿Ô±ÀîâÙ¡¢ÊµÏ°ÉúÀîã󣨱±¾©Óʵç´óѧÔÚ¶Á²©Ê¿Éú£©¡¢Ñо¿Ô±ê°Ã÷Ô¶ÒÔ¼°Öйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢Ìü¼¯ÍÅÊ×ϯ¿ÆÑ§¼Ò¡¢Öйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿ÔºÔº³¤Îâ½Ü½ÌÊÚÍê³É¡£¡£¡£¡£¡£¡£¡£¡£
Ä¿½ñ£¬£¬£¬£¬£¬£¬£¬£¬ÃæÏò¶à×â»§µÄGPU¼¯ÈºÒѳÉΪ´óÄ£×ÓѵÁ·£¨LLM£©µÄ½¹µã»ù´¡ÉèÊ©¡£¡£¡£¡£¡£¡£¡£¡£ÔÚ¶à¸öѵÁ·Ê¹Ãü¹²ÏíÍøÂç×ÊÔ´µÄÇéÐÎÏ£¬£¬£¬£¬£¬£¬£¬£¬ÓÉÓÚʹÃüÌØÕ÷±£´æ²î±ð£¬£¬£¬£¬£¬£¬£¬£¬ÏÖÓÐÖ÷Á÷ͨѶµ÷ÀíϵͳÄÑÒÔ°ü¹Ü¹«ÕýÐÔ£¬£¬£¬£¬£¬£¬£¬£¬³£µ¼Ö²¿·ÖʹÃü±»¡°¶öËÀ¡±»ò½ø¶ÈÖͺ󡣡£¡£¡£¡£¡£¡£¡£Í¨Ñ¶²»¹«Õý£¬£¬£¬£¬£¬£¬£¬£¬²»µ«Ó°Ïì¶à×â»§µÄÓû§ÌåÑ飬£¬£¬£¬£¬£¬£¬£¬ÖÆÔ¼¼¯ÈºµÄÕûÌåЧÄÜ£¬£¬£¬£¬£¬£¬£¬£¬¸üÖ±½ÓÍþвµ½ÖÇËãÔÆÐ§ÀͼòÖ±¶¨ÐÔ£¨Cloud Integrity£©ºÍÉÌÒµ×óȯ¡£¡£¡£¡£¡£¡£¡£¡£
Ϊ´Ë£¬£¬£¬£¬£¬£¬£¬£¬¸ÃÊÂÇéÁ¢ÒìÐÔÌá³ö¹éÒ»»¯½ø¶ÈÂÊ£¨Normalized Progress Rate£©Ö¸±ê£¬£¬£¬£¬£¬£¬£¬£¬Í¨¹ýȨºâʹÃüÔÚ¾ºÕùÇéÐÎϵÄÏÖʵ½ø¶ÈÓëÎÞ×ÌÈÅÀíÏë½ø¶ÈµÄ±ÈÀý£¬£¬£¬£¬£¬£¬£¬£¬¾«×¼Á¿»¯ÑµÁ·ÌåÑé¡£¡£¡£¡£¡£¡£¡£¡£ÕâһʹÃüÎ޹صÄÖ¸±êÀÖ³ÉÌî²¹Á˵ײãflow-level¹«ÕýÐÔÓëÉϲãÄ£×ÓѵÁ·job-level¹«ÕýÐÔÖ®¼äµÄÀíÂÛ¿Õȱ£¬£¬£¬£¬£¬£¬£¬£¬ÊǸÃÁìÓòµÄÖ÷ÒªÊÖÒÕÍ»ÆÆ£¬£¬£¬£¬£¬£¬£¬£¬»ò³ÉΪδÀ´ÐÐÒµ±ê×¼¡£¡£¡£¡£¡£¡£¡£¡£»£»£»£»£»£»£»£»ùÓÚ¸ÃÖ¸±ê£¬£¬£¬£¬£¬£¬£¬£¬Ñо¿ÍŶӹ¹½¨ÁËÍêÕûµÄ¹«ÕýÐÔÀíÂÛ£¬£¬£¬£¬£¬£¬£¬£¬²¢¿ª·¢ÁËLEVELLERϵͳ£¬£¬£¬£¬£¬£¬£¬£¬Ê×´ÎÔÚ¶à×â»§¼¯ÈºÖÐÕë¶Ôí§ÒâÊÂÇé¸ºÔØÊµÏÖͨѶµ÷ÀíµÄ×î´ó»¯-×îС»¯¹«Õý£¨Max-Min Fairness£©¡£¡£¡£¡£¡£¡£¡£¡£
LEVELLERϵͳ¼«¾ßÊÊÓÃÐÔÓë¿ÉÀ©Õ¹ÐÔ£¬£¬£¬£¬£¬£¬£¬£¬Ö§³ÖÔÚRDMAºÍTCPÏÖÓÐÓ²¼þÉÏÖ±½Ó°²ÅÅ¡£¡£¡£¡£¡£¡£¡£¡£ÊµÑéЧ¹ûÏÔʾ£¬£¬£¬£¬£¬£¬£¬£¬ÔÚ10ÖÖ´óÓïÑÔÄ£×ӵIJâÊÔÖУ¬£¬£¬£¬£¬£¬£¬£¬LEVELLERÏà±ÈÐÐÒµÖ÷Á÷¼Æ»®£¬£¬£¬£¬£¬£¬£¬£¬ÌáÉý×îµÍ½ø¶ÈÂÊ37%£¬£¬£¬£¬£¬£¬£¬£¬ÓÅ»¯¹«ÕýÐÔ17%£¬£¬£¬£¬£¬£¬£¬£¬Í¬Ê±¼á³Ö¼«¸ßµÄ¼¯Èº×ÊԴʹÓÃÂÊ¡£¡£¡£¡£¡£¡£¡£¡£¸ÃÊÂÇéΪ¶à×â»§AI¼¯ÈºÌṩÁËÐµĹ«ÕýÐÔ»ù×¼£¬£¬£¬£¬£¬£¬£¬£¬Ò²ÎªÖÇËãÖÐÐÄ£¨AIDC£©´ó¹æÄ£ÑµÁ·Í¨Ñ¶µ÷ÀíÌṩÇÐʵ¿ÉÐеĽâ¾ö¼Æ»®¡£¡£¡£¡£¡£¡£¡£¡£
±ðµÄ£¬£¬£¬£¬£¬£¬£¬£¬Öйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿Ôº³Â×ÓÐùÑо¿Ô±¼ÓÈëÍê³ÉµÄÏàÖúЧ¹û¡¶Scale-up
PIFO: Interleaving Multiple Priority Queues for High Speed Programmable
Scheduling¡·Ò²±»ACM SIGCOMM 2026ÈÎÃü¡£¡£¡£¡£¡£¡£¡£¡£¸ÃÊÂÇéÓɸ´µ©´óѧÐìÑï½ÌÊÚ¿ÎÌâ×éǣͷ£¬£¬£¬£¬£¬£¬£¬£¬ÃæÏòAIÊý¾ÝÖÐÐĺÍÐÂÐÍÔÆÍø»ù´¡ÉèÊ©Öн»Á÷»ú¶Ë¿ÚËÙÂÊÒ»Á¬ÌáÉý´øÀ´µÄ¸ßÐÔÄܵ÷ÀíÐèÇ󣬣¬£¬£¬£¬£¬£¬£¬Õë¶Ô¹Å°åµ¥PIFOÐÐÁÐÄÑÒÔÖ§³Ö1.6Tbps¼¶ÏßËÙ´¦Öóͷ£¡¢¼òÆÓ²¢Ðл¯ÓÖ»áÒýÈëµ÷ÀíÎó²îµÈÎÊÌ⣬£¬£¬£¬£¬£¬£¬£¬Ìá³ö¸ßËٿɱà³Ìµ÷Àí¿ò¼ÜScale-up PIFO¡£¡£¡£¡£¡£¡£¡£¡£¸Ã¿ò¼Üͨ¹ý½»Ö¯²¢Ðжà¸öPIFOÐÐÁÐÌáÉýµ÷ÀíÍÌÍ£¬£¬£¬£¬£¬£¬£¬£¬²¢Éè¼ÆRank Range Load BalancingËã·¨£¬£¬£¬£¬£¬£¬£¬£¬ÔÚ¿ØÖƵ÷ÀíÎó²îµÄͬʱ¼á³ÖÓ²¼þʵÏֵľ«Á·ÐÔ£¬£¬£¬£¬£¬£¬£¬£¬ÎªÏÂÒ»´ú¸ßËÙÊý¾ÝÖÐÐÄÍøÂçÖеĿɱà³ÌQoSµ÷ÀíÌṩÁËеÄÊÖÒÕ·¾¶¡£¡£¡£¡£¡£¡£¡£¡£
½üÄêÀ´£¬£¬£¬£¬£¬£¬£¬£¬ÑëÆóÔÚ»ù´¡Ñо¿ÓëÔʼÁ¢ÒìÁìÓòÒ»Á¬»ýÀÛ¡¢ºñ»ý±¡·¢¡£¡£¡£¡£¡£¡£¡£¡£Öйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿ÔºÑо¿ÍŶÓÔÚÖйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢Ìü¼¯ÍÅÊ×ϯ¿ÆÑ§¼Ò¡¢ÔÆÅÌËãÑо¿ÔºÔº³¤Îâ½Ü½ÌÊÚµÄÏòµ¼Ï£¬£¬£¬£¬£¬£¬£¬£¬Ò»Á¬Éî¸ûÔÆÅÌËãÍøÂç»ù´¡ÊÖÒÕÓëÒªº¦ÏÏû³Á¢Ò죬£¬£¬£¬£¬£¬£¬£¬´ÓÀíÂÛÌáÁ¶µ½ÏµÍ³ÊµÖ¤£¬£¬£¬£¬£¬£¬£¬£¬ÔÚÃæÏòÖÇÄÜÅÌËã»ù´¡ÉèÊ©µÄÍøÂçÒªº¦ÎÊÌâÉÏ¿ªÕ¹ºã¾Ã¹¥¹Ø¡£¡£¡£¡£¡£¡£¡£¡£´Ë´ÎЧ¹ûÈÎÃüACM SIGCOMM 2026£¬£¬£¬£¬£¬£¬£¬£¬ÌåÏÖÁËÖйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿ÔºÔÚ¹ú¼ÊÅÌËã»úÍøÂçÑо¿Ç°ÑصÄÔ´´Á¢ÒìÄÜÁ¦£¬£¬£¬£¬£¬£¬£¬£¬Ò²Åú×¢ÑëÆó²»µ«Äܹ»ÔÚÖØ´ó¹¤³Ì½¨ÉèÖС°¿¸´óÁº¡±£¬£¬£¬£¬£¬£¬£¬£¬ÕýÔÚ»ù´¡Ñо¿ÓëÔʼÁ¢ÒìÖÐÒ»Á¬·¢³öÖйúÆóÒµµÄÊÖÒÕÉùÒô¡£¡£¡£¡£¡£¡£¡£¡£
δÀ´£¬£¬£¬£¬£¬£¬£¬£¬Öйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿Ôº½«¼ÌÐø¼á³ÖÁ¢ÒìÇý¶¯Éú³¤£¬£¬£¬£¬£¬£¬£¬£¬Éî»¯ÔÆÅÌËãÍøÂç»ù´¡ÊÖÒսṹ£¬£¬£¬£¬£¬£¬£¬£¬Íƶ¯ÖØµã¿ÆÑÐЧ¹ûÏò½¹µãÊÖÒÕÄÜÁ¦×ª»¯£¬£¬£¬£¬£¬£¬£¬£¬²¢ÈÚÈëÌìÒíÔÆÆ½Ì¨ÄÜÁ¦ÏµÍ³£¬£¬£¬£¬£¬£¬£¬£¬Ò»Á¬ÔöǿҪº¦µ××ùÄÜÁ¦£¬£¬£¬£¬£¬£¬£¬£¬Ò»Ö±ÌáÉý×ÔÖ÷Á¢ÒìˮƽÓëϵͳ»¯¾ºÕùÓÅÊÆ¡£¡£¡£¡£¡£¡£¡£¡£Í¬Ê±£¬£¬£¬£¬£¬£¬£¬£¬Öйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢ÌüÔÆÅÌËãÑо¿Ôº½«½øÒ»²½Ê©Õ¹ÔÚÔÆÅÌËã¡¢ÍøÂçϵͳºÍÖÇÄÜÅÌËã»ù´¡ÉèÊ©ÁìÓòµÄÊÖÒÕ»ýÀÛÓëÈ˲ÅÓÅÊÆ£¬£¬£¬£¬£¬£¬£¬£¬ÎªÖйú¿·¢¡¤(Öйú)ÍøÕ¾-AGÆì½¢Ìü¡°ÔÆ¡ªÍø¡ªÊý¡ªÖÇ¡±ÈÚºÏÉú³¤ÌṩԽ·¢¼áʵµÄµ×²ãÊÖÒÕÖ§³Ö¡£¡£¡£¡£¡£¡£¡£¡£