TABLE II:

Dispatch, Alignment and Merge for BWA (Single End Reads)

# Partitions	Dispatch (seconds)						Alignment (seconds)						Merge (seconds)
	Non Secure			Secure			Non Secure			Secure			Non Secure	Secure
	min	avg	max	min	avg	max	min	avg	max	min	avg	max
10	14.45	14.52	14.45	44.00	44.58	44.87	0.27	0.43	0.77	5.19	7.58	8.70	0.88	4.70
20	6.57	7.81	8.06	22.35	26.62	27.26	0.12	0.14	0.16	2.32	4.10	4.77	0.83	4.65
40	3.20	4.46	4.56	9.63	14.32	14.58	0.06	0.10	0.14	1.05	1.77	2.15	0.80	4.62
80	1.48	1.48	2.93	6.29	9.12	9.68	0.03	0.13	0.47	0.44	0.73	1.18	0.81	4.66

The alignment step is parallelly executed across the cluster. Minimum, Average and Maximum time reported by containers.

Table II shows the results for the proposed partition and dispatch. The partition and dispatch approach shows a slowdown which greatly improves for higher partition counts due to better EPC utilization. Our approach makes it easier to run in parallel because of the pleasingly parallel nature of the data and the application. In the best case(if resources are available), we can run the single end alignment pipeline securely in 15.53 seconds (based on 9.68s in parallel dispatching, 1.18s in parallel alignment over 80 nodes and 4.66s in merging) by partitioning the problem into 80 subtasks. Even in the worst case that has only one SGX enabled node, we can expect to complete the alignment in 792.46 seconds running sequentially.