Performance of one 16-Gbit channel

The /390 server SE710 contains FC channels working at 16 Gbit/s. Here, the throughputs via one channel path with a cache hit rate of 100% are collocated.

The results for two cases are shown below:

• IOs/s with 4KB blocks and

• Mbyte/s when using 480-KB blocks.

FastDPAV-Aliases	Vol.	Tasks	seq. read		seq. write		random25
			IO/s (4 KB)	Mbyte/s (480 KB)	IO/s (4 KB)	Mbyte/s (480 KB)	IO/s (4 KB)	Mbyte/s (480 KB)
0	1	3	6,253	323	4,698	249	5,785	299
	1	8	6,255	323	4,543	81	5,781	299
	6	12	25,100	1,014	22,481	1,130	24,311	1,105
6	1	3	14,869	775	11,690	600	13,675	672
	1	8	24,688	1,012	21,443	402	23,776	1,103
	6	12	27,603	1,015	27,060	1,299	27,230	1,200

With 12 Tasks in parallel and using large IO block sizes, more than 1,000 Mbyte/s are achieved for read and up to 1,300 Mybte/s for write operation. Thereby, the limits of the IO system are reached.

If more than 3 tasks are working on 1 volume, waiting times (sw time) will sharply increase, leading to a lower overall throughput. Using FastDPAV, a significantly better throughput is achieved.

The tables below show the IO rates and throughputs for various block sizes for measdurements via 1 channel path, using 12 tasks on 6 volumes and FastDPAV.

Load type	IO rate in IO/s for 12 tasks with block size
	2 KB	4 KB	8 KB	16 KB	32 KB	160 KB	480 KB
sequential read	27,682	27,603	27,204	26,396	23,091	6,485	2,167
sequential write	27,150	27,060	26,791	26,299	24,899	7,849	2,772
random25	27,322	27,230	26,951	26,275	23,835	7,629	2,561

Load type	Throughput in Mbyte/s for12 tasks with block size
	2 KB	4 KB	8 KB	16 KB	32 KB	160 KB	480 KB
sequential read	54	107	212	412	721	1,013	1,015
sequential write	53	105	209	410	778	1,226	1,299
random25	53	106	210	410	744	1,192	1,200

For read IOs, maximum throughput is achieved with 128 KB and higher at ~1,000 Mbyte/s. For write IOs, it keeps rising with the block size.