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

Journal of Cloud Computing

Advances, Systems and Applications

Table 5 Percentages of reduced tail latency of DAG batch workloads driven by Alibaba trace 2018 under dynamic resources

Scenario	Job scheduling scenarios			TRP		BRA	EP	ILP	IPAP	LRP	MRP	NAP	NLP	NPAPP	RLP	TTP
17	DAG batch workloads	Daytime	Original resources	Restart rate	80%	18.10	20.17	26.92	18.80	17.03	13.64	22.13	15.56	21.49	19.49	21.16
18			50% decrease	Restart rate	80%	15.56	15.02	17.65	17.39	18.15	2.92	20.83	18.15	21.76	16.61	20.12
19			25% decrease	Restart rate	80%	19.08	24.29	17.19	25.35	21.48	8.23	26.39	27.89	20.30	28.14	22.63
20			25% increase	Restart rate	80%	16.99	26.92	25.00	26.29	28.45	13.64	28.15	32.68	16.59	34.48	25.33
21			50% increase	Restart rate	80%	18.78	31.03	28.89	29.20	30.13	14.89	25.93	34.96	27.93	33.88	29.52
22		Night	Original resources	Restart rate	80%	17.09	22.88	21.72	21.00	21.84	19.03	21.60	20.64	21.95	21.12	22.19
23			50% decrease	Restart rate	80%	16.28	10.05	10.48	9.17	15.52	0.85	9.26	11.92	10.05	7.64	9.96
24			25% decrease	Restart rate	80%	19.60	19.25	17.72	19.60	19.94	16.62	18.79	17.11	19.25	22.91	22.59
25			25% increase	Restart rate	80%	10.92	10.44	13.65	10.92	21.55	9.48	14.53	9.80	14.96	14.53	12.00
26			50% increase	Restart rate	80%	11.68	13.42	14.79	10.87	22.56	10.20	13.26	14.03	14.94	12.79	14.03