Fast DRL-based scheduler configuration tuning for reducing tail latency in edge-cloud jobs

Journal of Cloud Computing

Advances, Systems and Applications

Table 3 Percentages of reduced tail latency under AI workloads driven by Alibaba trace 2020

Scenario	Job scheduling scenarios			TDP/TRP		DRF _BRA	DRF _LRP	DRF _MRP	GANG _BRA	GANG _DRF _BINPACK	GANG _DRF _BRA	GANG _DRF _LRP	GANG _DRF _MRP	GANG _LRP	GANG _MRP	SLA _BRA	SLA _LRP	SLA _MRP
7	AI workloads	Daytime	Original resources	No policy	0%	34.99	29.41	23.95	28.49	26.74	33.04	20.00	32.36	35.48	22.58	34.14	27.62	28.32
8				Restart rate	80%	32.58	31.50	27.52	28.13	24.02	31.34	22.64	34.64	34.00	26.81	29.89	27.34	31.03
9				Restart rate	70%	27.90	26.39	18.55	22.47	17.66	22.77	17.99	29.93	29.43	19.54	24.91	22.37	26.57
10				Discard rate	80%	26.77	25.77	26.41	25.03	25.13	29.61	19.49	33.00	30.27	25.77	29.69	24.18	17.83
11				Discard rate	70%	16.94	15.42	15.77	15.06	15.30	21.83	8.00	25.25	21.73	15.18	20.79	14.08	10.61
12		Night	Original resources	No policy	0%	30.69	29.34	22.97	27.97	25.16	29.19	29.80	22.45	32.46	26.99	31.19	27.14	23.49
13				Restart rate	80%	25.32	23.27	19.51	22.61	23.58	25.05	26.95	18.67	23.06	21.33	26.79	21.76	21.25
14				Restart rate	70%	21.24	18.62	18.81	18.23	19.39	20.51	23.02	15.40	18.93	17.20	23.02	17.52	16.63
15				Discard rate	80%	19.04	17.15	16.95	16.18	16.18	20.96	15.93	14.54	17.15	20.10	20.48	15.22	19.12
16				Discard rate	70%	13.57	11.55	11.32	10.39	10.63	15.99	9.73	8.23	11.78	15.07	15.50	11.09	13.87